TW201947510A - Insurance service risk prediction processing method, device and processing equipment - Google Patents

Abstract

An insurance service risk prediction processing method, device and processing equipment: the calculation of a gradient boosting decision tree may be introduced into the prediction of insurance service risks, which not only is compatible with risk prediction processing for insurance service data of a non-linear relationship among insurance services, but may also output the relative risk size relationship after risk prediction, wherein a risk prediction result after sorting indicates the relative size of risk between different users, which may provide another more reliable implementation solution for predicting risk of insurance services.

Description

Processing method, device and processing equipment for insurance business risk prediction

The solutions in the embodiments of the present specification belong to the technical field of computer data processing for insurance business risk prediction, and more particularly, to a method, device, and processing device for insurance business risk prediction.

Motor vehicle insurance, or car insurance for short, refers to a type of commercial insurance that is liable for compensation for personal injury or property damage caused by a natural disaster or accident in a motor vehicle. With the development of the economy, the number of motor vehicles continues to increase. At present, auto insurance has become one of the largest types of insurance in China's property insurance business.
When a user insures a vehicle, an insurance company usually performs a risk assessment on the user. The result of the risk assessment will directly affect the user's insurance amount and preferential treatment. By assessing the risk of users, insurance companies can handle insurance business more accurately and reasonably, effectively avoiding or reducing business risks. At present, in the field of automobile insurance risk prediction, risk prediction based on a generalized linear model (GLM) has become the mainstream risk prediction technology system in the industry. The generalized linear model mainly deals with linearly related data objects. For example, the length of time spent on the Internet is reduced by 1 percentage point, and the age is increased by 1 year. GLM modeling can be realized based on the linear relationship between Internet age data and age data.
However, with the continuous increase of auto insurance business, the Internet data has shown a variety of types and massive data growth, and the traditional GLM model system has become more and more restricted. It is related to the shopping and habits of the crowd at the same time. The age distribution of different consumption habits changes with their own changes in a non-linear manner. The GLM model can summarize non-linear variables by binning, but it will lose the accuracy of many variables, making it difficult to adapt to current big data and multi-dimensional risk prediction requirements. Therefore, there is a great need in the industry for a processing method that can more effectively and efficiently perform risk prediction of auto insurance business in multi-dimensional data.

The purpose of the embodiments of the present specification is to provide a method, a device and a processing device for insurance business risk prediction, which can improve the decision tree by introducing a calculus gradient into the insurance business risk prediction, which is not only compatible with the insurance business data of non-linear relationships in the insurance business. The risk prediction process can also output the relative risk magnitude relationship after the risk forecast. The sorted risk prediction result represents the relative magnitude of the risk between different users, which can provide another more reliable risk prediction implementation solution for insurance business.
An insurance business risk prediction processing method, device, and processing equipment provided by the embodiments of this specification are implemented in the following ways:
A method for processing insurance business risk prediction, the method includes:
Obtain target risk related data of users to be predicted;
Use the constructed risk ranking model to process the target risk related data, and output the relative risk value of the user to be predicted, the risk ranking model includes: using the identified risk related data to train the calculus gradient promotion decision tree Determined sorting model.
A method for processing insurance business risk prediction includes:
Obtain target risk related data of users to be predicted;
The target risk related data is processed by using the constructed risk ranking model, and the relative risk value of the user to be predicted is output, and the relative risk value represents a relative risk magnitude relationship between users in a specified user set.
An insurance business risk prediction processing device includes:
Prediction data acquisition module, used to obtain target risk-related data of users to be predicted;
A risk prediction module is configured to process the target risk related data by using a constructed risk ranking model, and output the relative risk value of the user to be predicted. The risk ranking model includes: using the identified risk related data pair The calculus gradient boosts the decision tree for training to determine the ranking model.
An insurance business risk prediction processing device includes a processor and a memory for storing processor-executable instructions. When the processor executes the instructions, the processor implements:
Obtain target risk related data of users to be predicted;
Use the constructed risk ranking model to process the target risk related data, and output the relative risk value of the user to be predicted, the risk ranking model includes: using the identified risk related data to train the calculus gradient promotion decision tree Determined sorting model.
An insurance business risk prediction processing device includes a processor and a memory for storing processor-executable instructions. When the processor executes the instructions, the processor implements:
Obtain target risk related data of users to be predicted;
The target risk related data is processed by using the constructed risk ranking model, and the relative risk value of the user to be predicted is output, and the relative risk value represents a relative risk magnitude relationship between users in a specified user set.
An insurance business risk prediction processing method, device, and processing device provided in the embodiments of the present specification can obtain target risk-related data of users to be predicted, and then use the constructed risk ranking model to process the target risk-related data and output The relative risk value of the user to be predicted, and the risk ranking model includes a ranking model determined by training the calculus gradient promotion decision tree by using the identified risk-related data. Using the method provided by the embodiment of this specification, the decision tree can be improved by introducing a calculus gradient into the insurance business risk prediction, which is not only compatible with the risk prediction processing of insurance business data with non-linear relationships in the insurance business, but also can output the relative value after the risk prediction. The relationship between risk magnitudes, and the sorted risk prediction results represent the relative magnitude of risks between different users, which can provide another more reliable implementation of risk prediction for insurance business.

In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described in combination with the drawings in the embodiments of this specification. Obviously, the described The examples are only a part of examples in this specification, but not all examples. Based on one or more embodiments in this specification, all other embodiments obtained by a person having ordinary knowledge in the art without creative labor should fall within the protection scope of the embodiments of this specification.
With the development of computer Internet technology, the amount of data has grown rapidly. The classification of data characteristics in insurance business risk prediction is also more and more dimensional and detailed. The influence of many variables on screening classification is non-linear, such as the correlation between the length of time spent on the Internet and age, but the correlation can be diverse. For example, it can be a simple linear relationship, such as reducing the length of time spent online by 1 percentage point, and increasing age by 1 year; or a more complex relationship, such as an exponential relationship, reducing the length of time spent online by 4 percentage points and increasing age by 2 When it can be transformed into linearity through certain mathematical changes, it can be solved by generalized linear model. In real life, in addition to some variables with a basic linear relationship, there are also a large number of non-linear variables. For example, when predicting age, if "age" does not simply change with the length of time spent on the Internet, but is also related to the shopping and habits of the crowd, different consumption habits change the age distribution with its own changes in a non-linear manner. Because predicting the "user age" is one of the goals, if some linear relationship prediction models cannot identify non-linear relationships, it will greatly reduce the prediction performance of the model. In the existing solution, the variables can be summarized by sub-boxes, but the accuracy of many variables will be lost and the prediction result will be reduced. The embodiment of the present specification provides another method for realizing risk prediction in insurance business that is different from the existing conventional implementation, and introduces LambdaMART (Lambda Multiple Additive Regression Tree, λ-calculus gradient promotion decision tree, or λ-gradient promotion decision tree), Non-linear variables can be reasonably and effectively used in risk prediction to build a risk ranking model. The model is well compatible with linear and non-linear variables, and it can integrate risk relationships, and directly model the output of the relative risk of users in the same user set. It is as close to the real situation as possible, and the reliability of prediction results has been significantly improved. It should be noted that, for the convenience of description, LambdaMART may be referred to as a calculus gradient promotion decision tree in this specification.
The relative risk value described herein may include outputting a relative risk level relationship between a user in a user set and other users in a user set, and subsequent ones may be sorted according to specific values of the risk relative value. For example, in some embodiments of this specification, the relative risk of user A predicted by the LambdaMART model is 0.6, and the range of the relative risk value here is [0,1], which can indicate that the risk of the user A = 0.6 is greater than The risk of user B = 0.58. At this time, the value of 0.6 or 0.8 does not represent the specific payout rate or the absolute value of the predicted risk. It represents the relative risk of each user in some user sets. For example, in some application scenarios, although the risk of A is greater than the risk of B, but the actual payout rate of A and B is very low or the absolute value of the predicted risk is very low, both A and B may meet the risk assessment requirements based on the actual payout rate.
The user set described above can be specifically divided according to the processed data or the predicted demand of the actual application scenario. For example, users belonging to the same insurance company can be regarded as a user set, or users belonging to the same insurance type can be regarded as a user. Set, or all users responding to the constructed risk ranking model as a user set, or multiple users specified as a user set. Generally, a user set can be a user of the same insurance company or insurance service provider during model training, and it can be used without limitation in online prediction. For example, the relationship between the relative risks of multiple users of different insurance companies can be output.
In the following, a specific application scenario of risk prediction processing of auto insurance business is taken as an example to describe the implementation of this specification. Specifically, FIG. 1 is a schematic flowchart of an embodiment of a method for processing insurance business risk prediction provided in this specification. Although the present specification provides method operation steps or device structures as shown in the following embodiments or drawings, based on conventional or no creative labor, the method or device may include more or partially merged fewer operation steps. Or module unit. Among the steps or structures that do not logically have the necessary causal relationship, the execution order of these steps or the module structure of the device is not limited to the execution order or module structure shown in the embodiments or the drawings of this specification. When the described method or module structure is applied to an actual device, server, or end product, the method or module structure shown in the embodiment or the diagram may be executed sequentially or in parallel (for example, a parallel processor or Multi-line processing environment, even decentralized processing, server cluster implementation environment).
Of course, the description of the embodiment of the risk prediction of the automobile insurance business below does not limit other technical solutions that can be extended based on this specification. For example, in other implementation scenarios, the implementation solutions provided in this specification can also be applied to implementation scenarios such as fund risk ranking and medical insurance risk ranking. Applications in other implementation scenarios are described with reference to the embodiments of the automobile insurance business in this specification and will not be performed. Alternative repetitive description. A specific embodiment is shown in FIG. 1. An insurance business risk prediction processing method provided in this specification may include:
S0: Obtain target risk related data of users to be predicted;
S2: Use the constructed risk ranking model to process the target risk related data, and output the relative risk value of the user to be predicted. The risk ranking model includes: using the identified risk related data to improve the decision gradient of the calculation tree. Perform training to determine the ranking model.
The MART is another term for GBDT (Gradient Boosting Decision Tree). It is an iterative decision tree algorithm. The algorithm consists of multiple decision trees. The conclusions of all trees are added up to make the final answer. . The trees in GBDT are regression trees that can be used for regression prediction. In the insurance business risk prediction processing method provided in this specification, the underlying training model can use the GBDT non-linear relationship algorithm model, and the identified risk-related data can be used to construct a decision tree model in advance. ) Adjust and optimize the parameters in the decision tree step by step. When the model prediction result meets the accuracy requirements of insurance business risk prediction, it can be used online to predict the relative risk of the user to be predicted.
Lambda is the gradient used in the MART solution process. Its physical meaning can indicate the direction (up or down) and intensity of the next iteration of a file to be sorted. The core idea of GBDT is that in the continuous iteration, the regression decision tree model generated by the new iteration fits the gradient of the loss function, and finally all the regression decision trees are superimposed to obtain the final model. LambdaMART uses a special Lambda value to replace the above gradient, which is the sum of LambdaRank algorithm and MART algorithm. The combination of MART and Lambda can be used as the LambdaMART used in some embodiments of this specification. The principle of MART is to solve the function directly in the function space. The model result can be composed of many trees, and the fitting goal of each tree is the gradient of the loss function. LambdaMART's framework (the underlying model) is based on MART, which mainly uses Lambda for the gradients calculated in the middle. In LambdaMART, the parameters set by MART can include: the number of trees M, the number of leaf nodes L, and the learning rate v, etc. These parameters can be obtained by adjusting the validation set to obtain the optimal parameters.
MART supports "hot start", that is, you can continue training on the basis of the already trained model, and it can be loaded by initialization at the beginning. The following is a brief introduction to the general implementation of LambdaMART:
1. Each tree training will first traverse all the training data (file pairs with different labels), calculate the index changes and Lambda caused by the swap position of each pair, that is, then calculate the Lambda of each file, and then calculate each The derivative wi is used to solve the value of the leaf node in the Newton step.
2. Created in the first step of fitting the regression tree. The criterion for dividing the tree nodes is MSE (mean squared error, mean square error), and a regression tree with a number of leaf nodes is generated.
3. For the regression tree generated in the second step, calculate the value of each leaf node and use Newton step to solve, that is, for the file set that falls into the leaf node, use the formula to calculate the output value of the leaf node.
4. Update the model, add the currently learned regression tree to the existing model, and use the learning rate v (also called shrinkage coefficient) for regularization.
In a specific example, the algorithm process of LambdaMART can be shown in Figure 2, which mainly includes the following steps:
1) Determination of initial value. The model can be updated iteratively based on the initialized underlying model; if there is no underlying model, all initial values are 0;
2) Traverse the training set, calculate its lambda gradient (λ) for each file, and use it for subsequent Newton iterative methods to solve partial derivatives of leaf nodes;
3) Use the above gradient information to generate a decision tree. The criteria for dividing the tree nodes can refer to MSE to generate a regression tree R with leaf nodes L.
4) Calculate the value of the leaf nodes according to the Newton iteration method;
5) Update the model and update the score of each file according to the learning rate η.
This specification provides the lambdaMart used in the examples, which is different from the single user absolute correlation of the traditional regression model. The lambdaMart considers the comprehensive risk relationship of all users under a given condition and directly solves it. The result is more comprehensive. Generally, there is a certain difference in pricing between different insurance companies. When lambdaMart constructs training samples, it constructs an order relationship between the same insurance companies. The model training is based on the relative relationship between the data, not the absolute value. Unbalanced sample proportions are not sensitive. Relative to the absolute value of the regression model, this order relationship can more accurately represent the user's risk level. In addition, the lambdaMart in the embodiment of this specification uses a regression model of GBDT, which can perform risk prediction processing on the user characteristic data of the nonlinear relationship input by the model in the insurance business. It has strong applicability in the risk prediction of auto insurance users. The results are also more accurate and reliable.
In a specific implementation process, in one or more embodiments of the present specification, a risk ranking model based on LambdaMART may be constructed in advance. The training and construction of the specific GBDT model used at the bottom layer can be carried out according to the actual business scene requirements and information to set the corresponding model structure and parameter settings. For example, a single tree can be used for individual training, and the training residual can be used as the input of another tree to continue. Training; or multiple trees and multiple levels of connections for training, and the training residuals are then used as the input for another number of multiple levels of connections. Of course, in other embodiments, it is also possible to use the GBDT algorithm to perform some deformation, transformation or improvement of the non-linear relationship insurance business data risk prediction processing implemented by LambdaMART. This specification will not repeat the implementation process of LambdaMART model construction one by one. .
In this embodiment, training data of a risk ranking model may be determined in advance based on historical auto insurance business policy data collection, and training data may be identified according to risk division or setting requirements. In the implementation scenario of insurance business risk prediction in this embodiment, the training data may be referred to as risk-related data, and these risk-related data are usually associated with the insurance business and are used to train samples of the risk ranking model. For example, the risk correlation may be user characteristic data including multiple dimensions, the user characteristic data associated with one user is a set of training data, and each group of risk related data may be identified and set a corresponding risk score. Specifically, in an embodiment of the method described in this specification, the risk-related data may include user characteristic data related to at least one category, and the user characteristic data includes data information of a non-linear relationship associated with insurance business. For example, in one example, the risk-related data of the user A may include (A1, A2, A3, ..., A9) user characteristic data of 9 dimensions. User characteristics data in different dimensions can be selected accordingly according to the needs of auto insurance forecasting. For example, the 9 dimensions of the above example can include age, gender, occupation, annual income, historical outages, average monthly consumption, credit rating, marital status, and debt. assets. Alternatively, user characteristic data of 10 or more dimensions may be collected in advance, and user characteristic data for model training may be selected from user characteristic data of multiple dimensions when determining risk-related data. For example, specific risk-related information can include the following Table 1:



Of course, in other embodiments, the risk-related data may further include manual data generated according to a predetermined rule, for example, the operator may customize the risk-related data for model training according to the conditions that the expected risk may include. Or, after setting the data generation rules, the computer automatically generates the required risk-related data. The artificial data generated here is more in line with the expected risk prediction situation, while the historical auto insurance case data is closer to the real risk situation. In some implementation application scenarios, one of them can be used or a combination of artificial data and historical auto insurance case data can be used. Training of risk ranking models to improve the accuracy of prediction results.
The obtained risk correlation data can be used as training data for training in the GBDT model. After learning and training, the thresholds of decision characteristics (which can be all thresholds or part of the thresholds) when the decision tree branches in the risk ranking model can meet the model's finality. The relationship between the actual risk of each user identified (usually a continuous and stable output can also be required). The GBDT used in the embodiment of the present specification is an iterative decision tree algorithm, which can be mainly divided into a decision tree (Regression Decision Tree) and a gradient boosting (GB). Decision trees are mainly divided into two categories: classification trees and regression trees. Classification trees are commonly used to solve classification problems, such as user gender, whether a web page is a spam page, and whether a user cheats. The regression tree is generally used to predict real values, such as the age of the user, the probability of the user clicking, the relevance of the web page, and so on. The former is used to classify label paper, and the latter is used to predict real values. It should be emphasized here that the addition and subtraction of the results of the regression tree is meaningful, such as 10 years old + 5 years-3 years old = 12 years old, the latter is no way to accumulate or accumulate results is meaningless, such as male + male + female = male Is female. The general process of a regression tree is similar to a classification tree, except that each node of the regression tree will get a predicted value. Taking age as an example, the predicted value is equal to the average age of all people who belong to this node. When branching, each feature is exhaustively searched for the optimal segmentation variable and the optimal segmentation point. The criterion measured in this embodiment is no longer the Gini coefficient in the classification tree, but the square error is minimized. That is, the larger the number of people who are predicted to be wrong, the greater the squared error. The most reliable branch basis is found by minimizing the squared error. Branch until each person on the leaf node is interested in the game is unique or reaches a preset termination condition (such as the upper limit of the number of leaves). If the final age on the leaf node is not unique, everyone on the node The average age of is used as the prediction result of this leaf node.
Gradient boosting is a machine learning technique used for regression, classification, and ranking tasks, and is part of the Boosting algorithm family. Boosting is a family of algorithms that can promote weak learners to strong learners, and belongs to the category of ensemble learning. The Boosting method is based on the idea that, for a complex task, the judgment obtained by appropriately combining the judgments of multiple experts is better than the judgment of any one of the experts alone. In layman's terms, it is the principle of "three tanners better than one Zhuge Liang". Like other boosting methods, gradient boosting builds the final prediction model by ensemble multiple weak learners, usually decision trees. The boosting method uses a stage-wise way to build the model. The weak learner built at each step of the iteration is to make up for the shortcomings of the existing model.
For example, during a specific processing process, the tree of the tree can be set during training, and the tree of the tree can stop training after reaching a specified value (such as eighty); or the residual is small (meeting the conditions for stopping training) When these two conditions meet a training, you can stop training.
If the Nth residual is not all 0 or the stopping condition is not met, the residual results of the nodes of the Nth tree are used to replace the corresponding original values and substituted into the N + 1th tree for learning;
Until the residual value of the N + Kth number of leaf nodes is equal to or less than the threshold, the predicted value corresponding to the current leaf node is output. Specifically, all residuals can be accumulated as a predicted value.
In this embodiment, the number of decision trees used for training can be determined in advance, and the threshold of the decision characteristics when a decision tree is branched is gradually optimized through gradient iteration. For example, 80 decision trees can be used, and each tree learns the residuals of the sum of all previous tree conclusions. The threshold of the initial number can be set according to the experience value. Suppose that A's true score (the score for identification is 80), but the predicted score of the first tree based on the decision characteristics of age is 60 points, a difference of 20 points, and a residual error of 20. Then in the second tree (the decision characteristic is the user's occupation), set the score of A to 20 to learn. If the second tree can really score A to 20 leaf nodes, then add two trees. The conclusion is that A's true score (predicted score 60 points + residual 20 points); if the conclusion of the second tree is 18 points, then A still has 2 points of residuals, and the third tree (decision characteristic is year Income) A's age becomes 2 points, and continue to study. The residual calculation at each step is equivalent to increasing the weight of the mismatch event in disguise, and the time of pairing has tended to 0. For example, if the age is too old or too young, the greater the risk, and the higher the risk of income The smaller the age is, if a user is too old, but is classified into the less risky branch L1, but the average age of the less risky group L1 is between 20 and 40, the residual value will be It will increase accordingly, and the user can gradually divide it into the leaf nodes close to the actual risk through subsequent income, marital status, driving age, etc.
If the number of trained decision trees reaches a predetermined value, for example, all 10 trees from the root node to the leaf nodes are trained once, or the parameters of the current number meet the stopping training conditions, such as the residual error is 0 or other residual stopping thresholds, At this point, you can stop training for this group of data. When each threshold value finds the best segmentation point or the segmentation point that meets the training requirements, the threshold value of the decision feature of the decision tree can be determined, until the adjusted threshold value meets the prediction result output requirements of the risk ranking model, and the The risk ranking model is described. For example, a threshold of 60 and 80 for the risk score is initially set as whether the age is greater than 20 years. After training and optimizing with a large amount of data, the risk assessment from the age dimension can finally be used as a decision feature to adjust whether the age is greater than 24 years old to meet the true prediction results in most cases.
In another embodiment, the number of decision trees used by the risk ranking model may be determined based on the number of categories corresponding to the user characteristic data. For example, user characteristics data of 80 dimensions are selected, and each dimension can represent the decision characteristics of a tree. In this way, 80 decision trees can be used to build a non-linear risk ranking model. Of course, in other embodiments of the present specification, the total number of specific decision trees may be determined according to the collected data, the number of branches of the tree, the upper-lower connection relationship of the tree, and the like.
In an embodiment provided in this specification, the risk-related data used in the training of the risk ranking model is user characteristic data belonging to the same user set, and includes user characteristic data of at least one category, and the user characteristic data includes information related to insurance services. Information about connected nonlinear relationships.
The ranking model used in this embodiment. This model uses the compensation rates of users of the same insurance company to construct the relative risk relationship between these users during training. Since the user information in the same insurance company is completely comparable, this This guarantees that the training data is authentic and reliable. Therefore, compared with the traditional regression model, this embodiment provides another risk prediction recognition processing method based on the ranking model, and the results of the ranking model are more authentic and reliable. At the same time, the method provided by the embodiment of this specification can reasonably and effectively apply multi-dimensional nonlinear relationship variables in insurance business, and the risk ranking model based on the nonlinear relationship of the gradient promotion decision tree can be well compatible with linear and nonlinear variables. Compared with the traditional linear model, the accuracy of the prediction results has been significantly improved, which effectively makes up for the shortcomings of the traditional linear model, and improves the insurance business service experience. In another embodiment of this specification, specifically, the risk ranking model is in the training process The risk-related information used in this includes:
S20: Construct user feature data with the same user attribution label as a sample training set, and the user feature data includes data information of a non-linear relationship associated with the insurance service;
Correspondingly, performing the training includes: inputting the characteristic data of the first user in the first sample training set, and outputting the relative risk value of the first user in the first user set, and the first user set is The users included in the first sample training set are described.
The user attribution label may represent the distribution or classification of users. For example, a user of an insurance company may use a corresponding user attribution label to mark the user as belonging to the user attribution label of insurance company A.
The first sample training set, the first user set, and the first user described above mainly distinguish the currently processed training set from other training sets, and do not specifically refer to a certain set. By analogy, a training set composed of characteristic data of users of another insurance company B may be referred to as a second sample training set, and a set of users belonging to insurance company B may be referred to as a second user set. In model training, you can input the training set set constructed by each insurance company, and output the relative risk of the user to be predicted.
The risk ranking model can individually predict the relative risk value of each user, and can directly optimize the relationship between the relative risk of users belonging to the same user set that is output by the model. Therefore, in another embodiment of the method, the method may further include:
Determine the relative risk relationship between users in the specified user set based on the relative value of the risks;
Output data information of the relative risk magnitude relationship.
For example, in an example, the constructed risk ranking model can be used to process the characteristic data of users A, B, C, and D respectively. The relative risk values obtained are respectively 0.48, 0.56, 0.81, and 0.62, where B, C, and D belongs to the same insurance company. In some embodiments of the present specification, during the risk ranking model training phase, a sample training set is constructed by the characteristics of users of the same insurance company. Using such training samples for model learning and training can output more accurate and reliable Risk prediction results. Specifically, it can be output in this example, such as:
C (0.81)> D (0.62)> B (0.56)> A (0.48).
In the method provided in the embodiment of the present specification, a LambdaMART ranking model is used to input the user's characteristic data, and the numerical value of the user's payout rate can be output. The problem of poor reliability of the model output. LambdaMART's ranking model learns the level of risk between users. The values of the ranking model have no actual physical meaning and are used as a basis for comparing order relationships.
As mentioned above, the embodiments provided in this specification can be used not only in the implementation scenarios of risk prediction of auto insurance business, but also in the implementation scenarios of risk ranking of funds, ranking of medical insurance risks, and the like. Specifically in the application scenario of risk prediction for auto insurance business,
S22: The risk ranking model is a car insurance risk ranking model obtained by training based on risk-related data associated with the car insurance business;
The relative risk value includes a relative risk magnitude of a compensation rate corresponding to the user to be predicted.
Of course, the above-mentioned compensation rate and car insurance risk score are only an output characterization method of the non-linear relationship risk ranking model in one or more embodiments. This specification does not limit that other embodiments may also have other characterization methods or characterization methods in which the payout rate and car insurance risk score are transformed and transformed. For example, the car insurance score can be obtained after the payoff rate is linearly transformed. , The smaller the risk (opposite to the car insurance risk score, the greater the risk score, the higher the risk).
It should be noted that the linear relationship generally refers to the existence of a square function between two variables. The linear relationship of the variables in insurance or auto insurance described in the examples of this specification may include the form y = ax + b, where x is Independent variable, y is the controlled variable. In the specific application scenarios of the insurance or auto insurance business in the embodiments of this specification, the broad understanding of the linear relationship may mean that the relationship between two variables is clear and fixed. In some cases, it can be expressed in a straight line or by After a certain mathematical change, it is transformed into a linear relationship (the converted information loss is within a certain range). The non-linear relationship mainly refers to that the relationship between variables is constantly changing and cannot be described by formulas. In some cases, it can only be expressed by curves, surfaces or irregular lines, such as risk scores and occupations, risk scores Value and gender.
In one or more embodiments of the present specification, the process of constructing a risk ranking model may be generated in an offline pre-built manner, and training data including non-linear relationships may be selected in advance for learning training of the GBDT decision tree. After the training is completed, Use it online again. This specification does not exclude that the risk ranking model can be constructed or updated / maintained online. For example, if the computer capacity is sufficient, the risk ranking model can be constructed online, and the risk ranking model can be used simultaneously online and used for prediction. To process the target risk related data.
Although the above embodiment provides an implementation scheme that can implement a risk ranking model by using a LambdaMART model, this description does not exclude implementations in other embodiments that can use other data models to output relative risk magnitude relationships between users, such as other lambdarank ( (A sorting algorithm), listnet (a sorting algorithm), and other listwise methods (file list methods), as described above. Specifically, as shown in FIG. 3, this specification also provides another method for processing insurance business risk prediction. The method includes:
S30: Obtain target risk related data of users to be predicted;
S32: Use the constructed risk ranking model to process the target risk-related data, and output a relative risk value of the user to be predicted, where the relative risk value represents a relative risk magnitude relationship between users in a specified user set.
The relative risk output can be the value of one user or the value of multiple users. When outputting multiple user values, it can be the output result after sorting that represents the relative risk between users, or it can be the unsorted output result.
An embodiment of this specification provides a method for processing insurance business risk prediction, which can obtain target risk-related data of users to be predicted, and then use the constructed risk ranking model to process the target risk-related data to output the users to be predicted The relative risk value, the risk ranking model includes: a ranking model determined by training the calculus gradient promotion decision tree using the identified risk correlation data. Using the method provided by the embodiment of this specification, the decision tree can be improved by introducing a calculus gradient into the insurance business risk prediction, which is not only compatible with the risk prediction processing of insurance business data with non-linear relationships in the insurance business, but also can output the relative value after the risk prediction. The relationship between risk magnitudes, and the sorted risk prediction results represent the relative magnitude of risks between different users, which can provide another more reliable implementation of risk prediction for insurance business.
The method described above can be used for risk identification on the client side, such as risk assessment of insurance business provided in a payment application of a mobile terminal. The client can be a personal computer (PC), a server, an industrial control computer (industrial control computer), a mobile smart phone, a tablet electronic device, a portable computer (such as a notebook computer, etc.), and a personal digital assistant (PDA). , Or a desktop computer or smart wearable device. Mobile communication terminals, handheld devices, vehicle-mounted devices, wearable devices, TV devices, computing devices. It can also be applied to the system server of an insurance company or a third-party insurance service agency. The system server may include a separate server, a server cluster, a decentralized system server, or a server that processes data requested by the device and other Associated system server combination for data processing. For example, one implementation may include building on an Alibaba Cloud Open Data Processing Service (Open Data Processing Service, ODPS for short) platform. Can provide a unified programming interface and interface for a variety of data processing tasks from different user needs. System performance is guaranteed based on ODPS. The system implementing the method of the embodiment of this specification can process massive data in parallel and achieve the best computing performance.
As mentioned above, the method embodiments provided in the embodiments of this specification may be executed in a mobile terminal, a computer terminal, a server, or a similar computing device. Taking running on a server as an example, FIG. 4 is a block diagram of the hardware structure of a server applying the insurance business risk prediction processing method provided in this specification. As shown in FIG. 4, the server 10 may include one or more (only one shown in the figure) a processor 102 (the processor 102 may include, but is not limited to, a processing device such as a microprocessor MCU or a programmable logic device FPGA) , A memory 104 for storing data, and a transmission module 106 for communication functions. Those skilled in the art can understand that the structure shown in FIG. 4 is only a schematic diagram, and does not limit the structure of the electronic device. For example, the server 10 may further include more or fewer components than those shown in FIG. 4, for example, may further include other processing hardware, such as a database or a multi-level cache, or have a different configuration from that shown in FIG. Configuration.
The memory 104 may be used to store software programs and modules of application software, such as program instructions / modules corresponding to the search method in the embodiment of the present invention. The processor 102 runs the software programs and modules stored in the memory 104 , So as to execute various functional applications and data processing, that is, to achieve the above-mentioned processing method of navigation interactive interface content display. The memory 104 may include high-speed random access memory, and may further include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include memory remotely disposed relative to the processor 102, and these remote memories may be connected to the computer terminal 10 through a network. Examples of the above network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
The transmission module 106 is used to receive or send data through a network. The above specific examples of the network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission module 106 includes a network interface controller (NIC), which can be connected to other network devices through the base station to communicate with the Internet. In one example, the transmission module 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.
Based on the device model identification method described above, this specification also provides an insurance business risk prediction processing device. The device may include a system (including a decentralized system), software (application), a module, a component, a server, a client, and the like using the method described in the embodiments of the present specification, and a device incorporating necessary implementation hardware Device. Based on the same innovative concept, the processing device in one embodiment provided in this specification is as described in the following embodiments. Since the implementation solution of the device to solve the problem is similar to the method, the implementation of the specific processing device in the embodiment of this specification may refer to the implementation of the foregoing method, and the duplicated details are not described again. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and conceived. Specifically, as shown in FIG. 5, FIG. 5 is a schematic diagram of a module structure of an embodiment of an insurance business risk prediction processing device provided in this specification, which may include:
The prediction data acquisition module 201 may be used to obtain target risk-related data of a user to be predicted;
The risk prediction module 202 may be used to process the target risk correlation data by using the constructed risk ranking model, and output the relative risk value of the user to be predicted. The risk ranking model includes: using the identified risk correlation Data are used to determine the ranking model of calculus gradient promotion decision tree training.
It should be noted that, in the device described in the embodiments of this specification, according to the description of the related method embodiments, it may also include other implementations. For specific implementation manners, reference may be made to the description of the method embodiments, and details are not described herein.
The server or client provided in the embodiments of this specification can be implemented by the processor executing corresponding program instructions in the computer, such as using the C ++ language of the Windows operating system on the PC or server, or other systems such as Linux and the system. Corresponding application programming language sets the necessary hardware implementation, or the processing logic implementation based on quantum computers. The above processing equipment may specifically be an insurance server that provides risk prediction to an insurance server or a third-party service agency. The server may be a separate server, a server cluster, a distributed system server, or a processing equipment requesting data. Servers are combined with other system servers associated with data processing. This specification also provides an insurance business risk prediction processing device, which may specifically include a processor and a memory for storing processor-executable instructions. When the processor executes the instructions, it implements:
Obtain target risk related data of users to be predicted;
Use the constructed risk ranking model to process the target risk related data, and output the relative risk value of the user to be predicted, the risk ranking model includes: using the identified risk related data to train the calculus gradient promotion decision tree Determined sorting model.
Based on the embodiment of the foregoing manner, in another embodiment of the processing device provided in this specification, user characteristic data with the same user attribution label is constructed as a sample training set, and the user characteristic data includes information related to insurance services. Information about connected nonlinear relationships;
Correspondingly, performing the training includes: inputting the characteristic data of the first user in the first sample training set, and outputting the relative risk value of the first user in the first user set, and the first user set is The users included in the first sample training set are described.
Based on the foregoing embodiment, in another embodiment of the processing device provided in this specification, when the processor executes the instruction, it further implements:
Determine the relative risk relationship between users in the specified user set based on the relative value of the risks;
Output data information of the relative risk magnitude relationship.
Based on the embodiment of the foregoing manner, in another embodiment of the processing device provided in the present specification, the risk ranking model is an automobile insurance risk ranking model obtained by training based on risk-related data associated with a car insurance business;
The relative risk value includes a relative risk magnitude of a compensation rate corresponding to the user to be predicted.
Of course, the ranking model used in the risk prediction processing device provided in this specification is not limited to the LambdaMART model, and other algorithm models whose output indicates the relative magnitude of the user's risk can also be applied. Therefore, this specification also provides another insurance business risk prediction processing device, which may specifically include a processor and a memory for storing processor-executable instructions. When the processor executes the instructions, it implements:
Obtain target risk related data of users to be predicted;
The target risk related data is processed by using the constructed risk ranking model, and the relative risk value of the user to be predicted is output, and the relative risk value represents a relative risk magnitude relationship between users in a specified user set.
The above instructions can be stored in a variety of computer-readable storage media. The computer-readable storage medium may include a physical device for storing information, and the information may be digitized and then stored by using a medium such as electricity, magnetism, or optics. The computer-readable storage medium described in this embodiment may include: a device for storing information using electric energy, such as various types of memory, such as RAM, ROM, etc .; a device for storing information using magnetic energy, such as hard disk, floppy disk, Magnetic tape, core memory, bubble memory, USB flash drive; devices that use optical means to store information, such as CDs or DVDs. Of course, there are other ways of readable storage media, such as quantum memory, graphene memory, and so on. The instructions involved in the device or server or client or processing device described above are the same as described above.
It should be noted that the apparatus and processing equipment described in the embodiments of this specification may also include other implementation manners according to the description of the related method embodiments. For specific implementation manners, reference may be made to the description of the method embodiments, and details are not described herein.
Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the hardware + programming embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts may refer to the description of the method embodiment.
The specific embodiments of the present specification have been described above. Other embodiments are within the scope of the appended patent applications. In some cases, the actions or steps described in the scope of the patent application may be performed in a different order than in the embodiments and still achieve the desired result. In addition, the processes depicted in the figures do not necessarily require the particular order shown or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
An insurance business risk prediction processing method, device, and processing device provided in the embodiments of the present specification can obtain target risk-related data of users to be predicted, and then use the constructed risk ranking model to process the target risk-related data and output The relative risk value of the user to be predicted, and the risk ranking model includes a ranking model determined by training the calculus gradient promotion decision tree by using the identified risk-related data. Using the method provided by the embodiment of this specification, the decision tree can be improved by introducing a calculus gradient into the insurance business risk prediction, which is not only compatible with the risk prediction processing of insurance business data with non-linear relationships in the insurance business, but also can output the relative value The relationship between risk magnitudes, and the sorted risk prediction results represent the relative magnitude of risks between different users, which can provide another more reliable implementation of risk prediction for insurance business.
Although the present application provides method operation steps as described in the embodiments or flowcharts, more or less operation steps may be included based on conventional or non-creative labor. The sequence of steps listed in the embodiments is only one way of executing the steps, and does not represent the only sequence of execution. When the actual device or system server product is executed, it may be executed sequentially or in parallel according to the method shown in the embodiment or the diagram (for example, a parallel processor or a multi-line processing environment).
Although the content of the examples in this specification mentions the definition of linear / non-linear relationships, the construction of the GDBT underlying model in LambdaMART, the processing process of GBDT model algorithms, and other operations such as data acquisition, storage, interaction, calculation, and judgment, and The data description, however, the embodiments of the present specification are not limited to the situations described in the embodiments of the present specification that must conform to industry communication standards, standard GBDT model algorithm processing, communication protocols and standard data models / templates. Certain industry standards or implementations that are slightly modified based on implementations described in custom methods or embodiments can also achieve the same, equivalent or similar, or predictable implementation effects of the above embodiments. Embodiments obtained by applying these modified or deformed data acquisition, storage, judgment, processing methods, etc., may still fall within the scope of optional implementations of this specification.
In the 1990s, for a technical improvement, it can be clearly distinguished whether it is an improvement in hardware (for example, the improvement of circuit structures such as diodes, transistors, switches, etc.) or an improvement in software (for method and process Improve). However, with the development of technology, the improvement of many methods and processes can be regarded as a direct improvement of the hardware circuit structure. Designers almost always get the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that the improvement of a method flow cannot be realized by a hardware entity module. For example, a programmable logic device (
A Programmable Logic Device (PLD) (such as a Field Programmable Gate Array (FPGA)) is one such integrated circuit whose logic function is determined by the user programming the device. It is programmed by the designer to "integrate" a digital system on a PLD, without having to ask a chip manufacturer to design and manufacture a dedicated integrated circuit chip. Moreover, nowadays, instead of making integrated circuit chips by hand, this programming is mostly implemented using "logic compiler" software, which is similar to the software compiler used in program development and writing, and requires compilation. The previous source code must also be written in a specific programming language. This is called the Hardware Description Language (HDL). HDL is not only one, but there are many types, such as ABEL (Advanced
Boolean Expression Language), AHDL (Altera Hardware
Description Language), Confluence, CUPL (Cornell
University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. At present, the most commonly used is VHDL (Very-High-Speed
Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also be clear that the hardware circuit that implements the logic method flow can be easily obtained by simply programming the method flow into the integrated circuit with the above-mentioned several hardware description languages.
The controller may be implemented in any suitable manner, for example, the controller may take the form of a microprocessor or processor and a computer-readable storage of computer-readable code (such as software or firmware) executable by the (micro) processor. Media, logic gates, switches, application-specific integrated circuits (Application
Specific Integrated Circuit (ASIC), programmable logic controller and embedded microcontroller. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip
PIC18F26K20 and Silicone Labs C8051F320, the memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that, in addition to implementing the controller in pure computer-readable code, it is entirely possible to program the method steps to logically control the controller with logic gates, switches, application-specific integrated circuits, and programmable logic control. Controller and embedded microcontroller to achieve the same function. Therefore, the controller can be considered as a hardware component, and the device included in the controller for implementing various functions can also be considered as a structure in the hardware component. Or even, a device for implementing various functions can be regarded as a structure that can be both a software module implementing the method and a hardware component.
The processing equipment, devices, modules, or units described in the above embodiments may be specifically implemented by computer chips or entities, or by products with certain functions. A typical implementation is a computer. Specifically, the computer may be, for example, a personal computer, a notebook computer, a vehicle-mounted human-computer interactive device, a mobile phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, or a tablet. A computer, a wearable device, or a combination of any of these devices.
Although the embodiments of the present specification provide the operation steps of the method as described in the embodiments or flowcharts, more or less operation steps may be included based on conventional or non-creative means. The sequence of steps listed in the embodiments is only one way of executing the steps, and does not represent the only sequence of execution. When the actual device or terminal product is executed, it may be executed sequentially or in parallel according to the method shown in the embodiment or the diagram (for example, a parallel processor or multi-line processing environment, or even a distributed data processing environment). The terms "including,""including," or any other variation thereof are intended to encompass non-exclusive inclusion, such that a process, method, product, or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed Elements, or elements that are inherent to such a process, method, product, or device. Without further limitation, it does not exclude that there are other identical or equivalent elements in the process, method, product or equipment including the elements.
For the convenience of description, when describing the above device, the functions are divided into various modules and described separately. Of course, when implementing the embodiments of this specification, the functions of each module may be implemented in the same or multiple software and / or hardware, or the module that implements the same function may be composed of multiple submodules or subunits. Implementation etc. The device embodiments described above are only schematic. For example, the division of the unit is only a logical function division. In actual implementation, there may be another division manner. For example, multiple units or components may be combined or integrated into Another system, or some features, can be ignored or not implemented. In addition, the displayed or discussed mutual coupling or direct coupling or communication connection may be indirect coupling or communication connection through some interfaces, devices or units, which may be electrical, mechanical or other forms.
Those skilled in the art also know that, in addition to implementing the controller in pure computer-readable code, it is entirely possible to program the method steps to logically control the controller with logic gates, switches, application-specific integrated circuits, and programmable logic control. Controller and embedded microcontroller to achieve the same function. Therefore, such a controller can be considered as a hardware component, and the device included in the controller for implementing various functions can also be considered as a structure within the hardware component. Or even, a device for implementing various functions can be regarded as a structure that can be both a software module implementing the method and a hardware component.
The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and a combination of the flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, special purpose computer, embedded processor, or other programmable data processing device to generate a machine, so that the instructions generated by the processor of the computer or other programmable data processing device are generated A device for realizing the functions specified in one or more flowcharts and / or one or more blocks of the block diagram.
These computer program instructions may also be stored in computer readable memory that can guide a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory generate a manufactured article including a command device, The instruction device implements a function specified in a flowchart or a plurality of processes and / or a block or a block of a block diagram.
These computer program instructions can also be loaded on a computer or other programmable data processing equipment, so that a series of operating steps can be performed on the computer or other programmable equipment to generate computer-implemented processing, which can be executed on the computer or other programmable equipment. The instructions provide steps for implementing the functions specified in one or more flowcharts and / or one or more blocks of the block diagram.
In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.
Memory may include non-permanent memory, random access memory (RAM), and / or non-volatile memory in computer-readable media, such as read-only memory (ROM) or flash memory ( flash RAM). Memory is an example of a computer-readable medium.
Computer-readable media includes permanent and non-permanent, removable and non-removable media. Information can be stored by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), and other types of random access memory (RAM) , Read-only memory (ROM), electrically erasable and programmable read-only memory (
EEPROM), flash memory or other memory technology, CD-ROM, digital versatile disk (DVD) or other optical storage, magnetic tape cartridges, magnetic disk storage or other magnetic storage devices Or any other non-transmitting medium that can be used to store information that can be accessed by a computing device. According to the definition in this article, computer-readable media does not include temporary computer-readable media (transitory media), such as modulated data signals and carrier waves.
Those skilled in the art should understand that the embodiments of the present specification may be provided as a method, a system, or a computer program product. Therefore, the embodiments of the present specification may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the embodiments of the present specification may use a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) containing computer-usable code. form.
The embodiments of this specification can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The embodiments of the present specification can also be practiced in a decentralized computing environment. In these decentralized computing environments, tasks are performed by a remote processing device connected through a communication network. In a decentralized computing environment, program modules can be located in local and remote computer storage media, including storage devices.
Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple. For the relevant part, refer to the description of the method embodiment. In the description of this specification, the description with reference to the terms “one embodiment”, “some embodiments”, “examples”, “specific examples”, or “some examples” and the like means specific features described in conjunction with the embodiments or examples , Structure, materials, or features are included in at least one embodiment or example of an embodiment of the present specification. In this specification, the schematic expressions of the above terms are not necessarily directed to the same embodiment or example. Moreover, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. In addition, without any contradiction, those skilled in the art may combine and combine different embodiments or examples and features of the different embodiments or examples described in this specification.
The above descriptions are merely examples of the embodiments of the present specification, and are not intended to limit the embodiments of the present specification. For those skilled in the art, the embodiments of the present specification may have various modifications and changes. Any modification, equivalent replacement, and improvement made within the spirit and principle of the embodiments of the present specification shall be included in the scope of patent application of the embodiments of the present specification.

S30 ~ S32‧‧‧step

10‧‧‧Server (computer terminal)

102‧‧‧ processor

104‧‧‧Non-volatile memory

106‧‧‧Transmission Module

201‧‧‧ Forecast data acquisition module

202‧‧‧Risk Forecast Module

In order to more clearly explain the embodiments of the present specification or the technical solutions in the prior art, the drawings used in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only For some people with ordinary knowledge in the field, some of the embodiments described in the specification can also obtain other drawings based on these drawings without paying creative labor.

FIG. 1 is a schematic flowchart of an embodiment of an insurance business risk prediction processing method provided in this specification;

2 is a schematic diagram of a LambdaMART model processing process in the method provided in this specification;

3 is a schematic flowchart of another embodiment of an insurance business risk prediction processing method provided in this specification;

FIG. 4 is a block diagram of the hardware structure of a server to which an insurance business risk prediction processing method provided in this specification is applied.

FIG. 5 is a schematic diagram of a module structure of an insurance business risk prediction processing device provided in this specification.

Claims (11)

A method for processing insurance business risk prediction, the method includes: Obtain target risk related data of users to be predicted; Use the constructed risk ranking model to process the target risk related data, and output the relative risk value of the user to be predicted. The risk ranking model includes: using the identified risk related data to train and determine the gradual decision tree for gradual improvement. model. According to the method described in the first patent application scope, the risk correlation data used in the training process of the risk ranking model includes: Constructing user characteristic data with the same user attribution label as a sample training set, the user characteristic data including non-linear relationship data information associated with insurance business; Correspondingly, the training includes: inputting the characteristic information of the first user in the first sample training set, and outputting the relative risk value of the first user in the first user set, and the first user set Users included in this training set. The method according to item 2 of the patent application scope, further comprising: Based on the relative risk value, determine the relative risk magnitude relationship between users in the specified user set; Output data information of the relative risk magnitude relationship. As in the method described in item 3 of the scope of patent application, the risk ranking model is an auto insurance risk ranking model trained based on risk-related data associated with the car insurance business; The relative risk value includes the relative risk magnitude of the compensation rate corresponding to the user to be predicted. A method for processing insurance business risk prediction includes: Obtain target risk related data of users to be predicted; The target risk related data is processed by using the constructed risk ranking model, and the relative risk value of the user to be predicted is output. The relative risk value represents the relationship between the relative risks among users in the specified user set. An insurance business risk prediction processing device includes: Prediction data acquisition module, used to obtain target risk-related data of users to be predicted; The risk prediction module is used to process the target risk related data by using the constructed risk ranking model and output the relative risk value of the user to be predicted. The risk ranking model includes: using the identified risk related data to improve the calculation gradient The decision tree is trained to determine the ranking model. An insurance business risk prediction processing device includes a processor and a memory for storing processor-executable instructions. When the processor executes the instructions, the processor implements: Obtain target risk related data of users to be predicted; Use the constructed risk ranking model to process the target risk related data, and output the relative risk value of the user to be predicted. The risk ranking model includes: using the identified risk related data to train and determine the gradual decision tree for gradual improvement. model. As the processing equipment described in item 7 of the scope of patent application, the risk-related data used in the training process of the risk ranking model includes: Constructing user characteristic data with the same user attribution label as a sample training set, the user characteristic data including non-linear relationship data information associated with insurance business; Correspondingly, the training includes: inputting the characteristic information of the first user in the first sample training set, and outputting the relative risk value of the first user in the first user set, and the first user set is the same as the first user Users included in this training set. As the processing device described in item 8 of the patent application scope, when the processor executes the instruction, it also implements: Determine the relative risk relationship between users in the specified user set based on the relative value of the risks; Output data information of the relative risk magnitude relationship. According to the processing equipment described in item 9 of the scope of patent application, the risk ranking model is an auto insurance risk ranking model obtained by training based on risk-related data associated with auto insurance business; The relative risk value includes the relative risk magnitude of the compensation rate corresponding to the user to be predicted. An insurance business risk prediction processing device includes a processor and a memory for storing processor-executable instructions. When the processor executes the instructions, the processor implements: Obtain target risk related data of users to be predicted; The target risk related data is processed by using the constructed risk ranking model, and the relative risk value of the user to be predicted is output. The relative risk value represents the relative risk magnitude relationship between users in the specified user set.