CN113837873B - Method, equipment and computing medium for estimating waiting time of automobile financial loan - Google Patents
Method, equipment and computing medium for estimating waiting time of automobile financial loan Download PDFInfo
- Publication number
- CN113837873B CN113837873B CN202111220745.5A CN202111220745A CN113837873B CN 113837873 B CN113837873 B CN 113837873B CN 202111220745 A CN202111220745 A CN 202111220745A CN 113837873 B CN113837873 B CN 113837873B
- Authority
- CN
- China
- Prior art keywords
- time
- application form
- neural network
- queue
- application
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/03—Credit; Loans; Processing thereof
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/25—Integrating or interfacing systems involving database management systems
- G06F16/254—Extract, transform and load [ETL] procedures, e.g. ETL data flows in data warehouses
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Abstract
The method comprises a data acquisition and processing stage, a model training stage and a prediction stage, wherein an automobile finance loan application form is firstly acquired, characteristic information is extracted and then encoded, and the characteristic information comprises time characteristics, risk level characteristics and quantity related characteristics; then constructing a neural network model, and training and verifying the neural network model by using a sample set, wherein the sample set comprises a coded value of characteristic information corresponding to each automobile financial loan application form and actual waiting time, which are obtained in a data acquisition and processing stage; and finally, acquiring an automobile financial loan application form to be predicted, extracting corresponding characteristic information codes, inputting the characteristic information codes into a trained neural network model, and outputting a predicted waiting time result by the trained neural network model. Compared with the traditional isochronous prediction scheme, the method and the device can process more complex data, obtain more accurate prediction time and realize more automatic model deployment.
Description
Technical Field
The invention belongs to the technical field of queuing prediction, and relates to a neural network-based prediction method, equipment and a computing medium for the waiting time of an automobile financial loan.
Background
When a customer waits for the approval of an automobile financial loan, the waiting time of the uncertain time is longer than the limited waiting time of the known time, the waiting time prediction model can provide accurate predicted time, so that the customer can know the loan progress and time, the customer can conveniently confirm the ordering vehicle arrangement, the dissatisfaction caused by the waiting time is relieved, and the experience of the customer is improved as much as possible.
At present, related technologies are proposed in the fields of prediction of restaurant number calling and the like, prediction of takeaway waiting time and the like, but the following disadvantages still exist:
(1) Because the types of data processed in different fields are different and the isochronous rules are different, some waiting time estimation schemes with good effects in other fields are not suitable for the automotive financial business scene: if the take-out waiting time is up, the customer already determines the restaurant after finishing the selection, which is equivalent to that each restaurant is an independent queue; one application form in the automobile financial business scene cannot determine auditors before processing, because one auditor can process application forms with multiple risk levels; for another example, the travel waiting time is estimated based on the number of vehicles, distance and weather factors, and the processing logic of the selected data type does not conform to the business logic of the automobile financial business scene.
(2) The traditional statistical model calculates the average processing time of 1 application form per time period (10 minutes) and then adds up the accumulations before the current application form to get the waiting time. The scheme is suitable for scenes with small fluctuation of the number of queuing people along with time, and automobile loans are influenced by factors such as new automobile release, holidays, automobile exhibition and the like, the real-time application data change of the current day cannot be utilized, and the prediction accuracy is poor.
(3) Conventional queuing follows the first come, but in the automotive financial business scenario, the emergency (queue insertion) situation frequently occurs due to special situations (such as supplementary traffic data, customer reasons and the like), and the related waiting time prediction scheme supports less emergency situations. In addition, the processing time required by the loan auditing individuals is greatly different, and the waiting time can be influenced by other application forms with different levels, which all have difficulty in accurately predicting the time.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides a neural network-based prediction method, equipment and a computing medium for the waiting time of an automobile finance loan, a data processing scheme and a model building scheme are designed for the data type related to the field of the automobile finance loan, and the automatic model training and deployment can be realized through codes, so that the prediction accuracy is improved.
In order to achieve the above purpose, the technical scheme adopted by the invention is as follows:
a method for predicting the waiting time of an automotive financial loan, the waiting time predicting method comprising:
data acquisition and processing stage: collecting a plurality of automobile financial loan application forms, extracting and encoding characteristic information of each automobile financial loan application form, wherein the characteristic information comprises time characteristics, risk level characteristics and quantity-related characteristics, and the quantity-related characteristics comprise the accumulation quantity, the checking quantity and the urgent quantity of the application forms in the corresponding risk level of each automobile financial loan application form when entering a waiting queue;
model training stage: constructing a neural network model, and training and verifying the neural network model by utilizing a sample set, wherein the sample set comprises a coded value of characteristic information corresponding to each automobile financial loan application form and actual waiting time, which are obtained in the data acquisition and processing stage; the neural network model takes the coding value of the characteristic information of the automotive financial loan application form as input and takes the actual waiting time as output as a target;
prediction stage: and acquiring an automobile financial loan application form to be predicted, extracting corresponding characteristic information, encoding, inputting the encoded characteristic information into a trained neural network model, and outputting a predicted waiting time result by the trained neural network model.
Further, the temporal features extracted during the data acquisition and processing stage are represented by (a, b, c), wherein a, b, and c are positive integers, and a e [1,7] represents monday through sunday, respectively; dividing 24 hours a day into a time period every 30 minutes, b epsilon [0, 47] indicating that b time periods of the day have elapsed; c.epsilon.0, 29 represents the c-th minute in the b+1 time period.
Further, in the data acquisition and processing stage, the time feature is subjected to mean value coding to obtain a coded value of the time feature, and the formula is as follows:
wherein x is the time characteristic to be encoded, y is the target value, n i Is x=x i The number of samples in the time, N is the total number of samples, the coefficient
Further, in the data acquisition and processing stage, onehot encoding is performed on the risk level features to obtain encoded values of the risk level features.
Further, the time characteristics extracted by the data acquisition and processing stage include an auto finance loan application form in-queue time, a processing start time, and a processing completion time.
Further, in the data collection and processing stage, the method for calculating the accumulation number of the application forms in the risk level corresponding to any application form when entering the waiting queue comprises the following steps: counting the number of the application forms, of which the processing completion time is later than the queue entering time of the application form apply_1 and the queue entering time is earlier than the queue entering time of the application form apply_1, in the application form which needs to be calculated is apply_1 and has the same risk level as the application form apply_1;
the method for calculating the auditing quantity of the application forms in the corresponding risk level when the application form application_1 enters the waiting queue comprises the following steps: counting the number of the application forms with the same risk level as the application form apply_1, wherein the processing completion time in the application form apply_1 is later than the queue entering time of the application form apply_1, and the processing starting time is earlier than the queue entering time of the application form apply_1;
the method for calculating the urgent number of the application forms in the corresponding risk level when the application form application_1 enters the waiting queue comprises the following steps: counting the number of the application forms with the same risk level as the application form apply_1, wherein the time of entering the queue in the application form apply_1 is later than the time of entering the queue in the application form apply_1, and the processing starting time is earlier than the processing starting time of the application form apply_1.
Further, in the data acquisition and processing stage, the characteristic information of each automobile financial loan application form is normalized, and the coding parameters and the normalization parameters are stored in a database, wherein the data in the sample set are normalized data.
Further, in the model training stage, a 3-layer fully connected neural network is constructed by using tensorsurface, and an activation function is a relu function; and inputting the sample set into a neural network, and continuously adjusting neural network parameters through back propagation to fit a label to obtain a final trained neural network model, wherein the label is the actual waiting time of each automobile financial loan application form in the sample set.
Further, the neural network model includes tree model-based lightgbm and xgboost.
Further, in the model training stage, after each training set time, calculating an error value of the waiting time predicted according to the current neural network model and the actual waiting time of the corresponding application form as a current error; comparing the current error with a set reference value, if the current error is smaller than the reference value, comparing the current error with a minimum error value obtained in the previous training, and if the current error is smaller, storing the weight of the current neural network model, and continuing training; if the current error is not smaller than the reference value or is not smaller than the minimum error value obtained in the previous training, the weight of the current neural network model is not saved, and the training is continued; and terminating the training when the current errors in the multiple comparisons are not smaller than the reference value or are not smaller than the minimum error value obtained in the previous training.
Further, in the data acquisition and processing stage, firstly, automatically pulling data by using an ELT script, wherein ELT is the processing capability of a database, E=extracting data from a source database, L=loading the data into a temporary table of a target database, T=converting the data in the temporary table and then loading the data into a target table of the target database; then, data cleaning and characteristic engineering are carried out, and the obtained coding parameters, coding rules and normalized parameters after coding and normalizing the characteristic information are stored as model deployment data; and in the model training stage, carrying out model deployment on the trained neural network model, in the prediction stage, inputting the characteristic information corresponding to the automotive financial loan application form to be predicted into the deployed neural network model, and storing the obtained prediction waiting time result into a database.
The invention also provides equipment for realizing the method for estimating the waiting time of the automobile financial loan, which comprises the following steps:
a memory for storing a computer program;
and the processor is used for realizing the method for estimating the waiting time of the automobile financial loan when executing the computer program.
The invention also provides a computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of the method of predicting a financial loan wait time for an automobile as described above.
The invention has the beneficial effects that:
according to the data characteristics in the automotive finance field, the invention designs a scheme capable of realizing waiting time estimation, and compared with the traditional isochronous estimation scheme, the scheme can process more complex data types, can adapt to real-time change of data, and can process application forms with different individual levels.
The prediction accuracy is improved, experiments show that the average absolute error of the prediction is 3.8 minutes, the accuracy of the error within 4 minutes reaches 96.2%, and the prediction accuracy is more and more accurate along with the increase of data.
In addition, the invention can realize automatic model training and deployment through codes, reduce the burden of operators and avoid causing data errors.
Drawings
Fig. 1 is a schematic diagram of a feasible flow for implementing data acquisition and processing stages in an automotive finance loan waiting time estimation method based on a neural network.
Fig. 2 is a schematic diagram of a feasible flow chart for implementing a model training stage in an automobile finance loan waiting time estimation method based on a neural network.
Fig. 3 is a schematic diagram of a neural network connection used in a method for estimating the waiting time of an automotive finance loan based on a neural network according to the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. Embodiments of the present invention are intended to be within the scope of the present invention as defined by the appended claims. It is to be understood that the terminology used in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.
It should be understood that the step numbers used herein are for convenience of description only and are not limiting as to the order in which the steps are performed. As used in this specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. The terms "comprises" and "comprising" indicate the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.
Example 1
The embodiment provides an automobile finance loan waiting time estimation method based on a neural network, which comprises a data acquisition and processing stage, a model training stage and a prediction stage.
In the data acquisition and processing stage, a plurality of automobile finance loan application forms are acquired first, and the characteristic information of each automobile finance loan application form is extracted and encoded. According to the embodiment, queuing time related data including policy data, personnel data, flow time node data and the like are collected at a big data platform buried point for about half a year; specifically, extracting data from a large data platform forms mainly 4 tables: the application form processing flow chart, the application form grade rule chart and the approver grade rule chart are connected through simple data processing such as a chart, and the final production field comprises: the fields can be extended as needed, and in this embodiment, the fields are handled by taking the above fields as an example, where the number of the application form, the order of the application form, the time of entering the queue, the processing start time, the processing completion time, the approver level, the urgent mark, and the like are about 40 ten thousand.
The application form level may include a risk level, and the business system automatically classifies and determines the corresponding risk level according to information (such as credit records before the applicant) in the application form, which is a risk level for representing credit, and may be classified into ten risk levels of "a", "AA", "AAA", "B", "BB", "BBB", "C", "CC", "CCC" and "D".
According to the loan approval process, factors with great influence on queuing time are required to be extracted, and the characteristic information of each automobile financial loan application form extracted by the embodiment comprises time characteristics, risk level characteristics and quantity related characteristics. The number-related features include the accumulated number of application forms, the checking number and the urgent number in the risk level corresponding to each automotive finance loan application form when entering the waiting queue, and the number-related features are 3×10 fields in total because the risk level is classified into ten levels in the embodiment.
In the model training stage, a neural network model is built and is trained and verified by using a sample set, wherein the sample set comprises characteristic information (comprising time characteristics, risk level characteristics and quantity related characteristics) corresponding to each automobile financial loan application form and actual waiting time, which are obtained in the data acquisition and processing stage, the characteristic information is used as the input of the neural network model, the output of the neural network model is predicted waiting time, and parameters and weights of the model are optimized through continuous training and verification, so that the predicted waiting time output by the neural network model is as close to the actual waiting time as possible. The sample set is typically required to be divided into a training set and a test set for use in model training and testing, and this embodiment is performed according to 8:1:1 divide the sample set into a training set, a validation set and a test set.
After the model is trained, a prediction stage is entered, firstly, an automobile financial loan application form to be predicted is obtained, then, characteristic information corresponding to the application form is extracted so that the application form is converted into a form meeting the input requirement of the model, the extracted characteristic information is input into a trained neural network model, and the trained neural network model can output a predicted waiting time result.
Example two
The present embodiment is modified on the basis of the first embodiment to provide a specific method for extracting temporal features, but it is obvious that the temporal feature extraction method of the present invention is not limited to this form of the present embodiment.
The time feature may include an in-queue time, a processing start time and a processing completion time of the automotive financial loan application form, taking the extraction of the in-queue time as an example, the elements of the time feature are mainly the time, day of week and time period of the application form in which the application form is in the queue, the embodiment considers that business on-duty and off-time (such as 8.30 on-duty and 11.30-1.30 on noon) are divided into a time period in 30 minutes, the form of which the time period is in the time period is expressed, and the expression form of the extracted time feature is (a, b, c), wherein a, b and c are positive integers, and a e [1,7] respectively represents monday to sunday; dividing 24 hours a day into a time period every 30 minutes, b epsilon [0, 47] indicating that b time periods of the day have elapsed; c.epsilon.0, 29 represents the c-th minute in the b+1 time period. The post-time extraction performance form is (3, 21, 13) as 2021-08-25 10:43:30, the first 3 representing 2021-08-25 being wednesday, the second 21 and third 13 representing 10:43 passing 21 time periods of the day, 13 minutes in the 22 th time period.
It was verified that using this method, the accuracy of the error within 4 minutes was improved from 92.1% to 96.2% relative to the number of hours of direct use as a feature.
Example III
The embodiment improves on the basis of the first embodiment and the second embodiment, and provides a specific method for coding the time characteristics, wherein the embodiment codes the time characteristics by the mean value, and the formula is as follows:
where x is the temporal feature to be encoded, such asWhen encoding the extracted temporal feature of embodiment two, x refers to three fields a, b, and c, and xi represents a temporal feature of a certain actual value, such as (3, 21, 13) in embodiment two. n is n i Is x=x i The number of samples in the time, N is the total number of samples, y is the target value,
representing x=x i Mean value of y corresponding to time, +.>
Is the average value of y over the entire training set.
λ(n i ) Represents the balance coefficient for the specific case and the whole, lambda (n i )∈[0，1]Responsible for calculating the reliability of the two probability values, λ (n i ) =0.5 means that the reliability of the two probabilities is equal, with n i The reliability of the prior probability gradually decreases. If a new feature class appears in the test set (not in the training set, e.g., the training set is typically on duty, if the test set is handled by a salesman after off duty, and then the new feature class appears in the test set), λ (n i ) =1, which indicates that the average value of the history is directly used to infer a value that has not occurred. In this embodiment, the coefficient λ (n) is finally characterized by the following formula after business analysis and last 100 experiments i )：
Wherein n is i For x=x i For a total of 3 fields. This embodiment improves the accuracy of the error within 4 minutes from 89.3% to 96.2% relative to a scheme that does not use mean value coding.
In addition, the risk level features are also required to be encoded, in this embodiment, the onehot encoding mode is adopted, and the onehot encoding mode is also called one-bit effective encoding, mainly adopts a multi-bit state register to encode a plurality of corresponding states, and each state has an independent register bit and is only one bit effective at any time. For example, in the first embodiment, risk grades are classified into ten risk grades of "a", "AA", "AAA", "B", "BB", "BBB", "C", "CC", "CCC" and "D", and an application form with risk grade a can be converted into [1,0,0,0,0,0,0,0,0,0] according to onehot code, and total 10 fields are included; the application form with the risk grade of AA can be converted into [0,1,0,0,0,0,0,0,0,0] according to onehot codes; and so on.
In the data acquisition and processing stage, an approver grade rule table is extracted from a big data platform, the grade of the approver can be divided into multiple grades, and each grade of approver can process application forms with different risk grades, so that the application forms with proper risk grades and proper quantity can be distributed to the approver by combining the risk grade setting scheme in actual approval, and the approval efficiency is improved.
Example IV
The embodiment is improved on the basis of the first to third embodiments, the manner of calculating the related characteristics of the quantity from the application forms is limited, during the data processing, all the application forms are ordered according to the time of entering the queue, the application forms are sequentially selected, the currently selected application form is recorded as application_1, the processing completion time is reserved to be earlier than the processing start time of application_1, and the processing start time is later than the time of entering the queue of application_1. And traversing all application forms in sequence, and respectively counting 3 data, namely the accumulated quantity, the checking quantity and the urgent quantity of the application forms in the corresponding risk level when each automobile finance loan application form enters a waiting queue.
The method for calculating the accumulation number of the application forms in the corresponding risk level when the application form application_1 enters the waiting queue comprises the following steps: counting the number of the application forms with the same risk level as the application form apply_1, wherein the processing completion time in the application form apply_1 is later than the entering queue time of the application form apply_1, and the entering queue time is earlier than the entering queue time of the application form apply_1.
The method for calculating the auditing quantity of the application forms in the corresponding risk level when the application form application_1 enters the waiting queue comprises the following steps: counting the number of the application forms with the same risk level as the application form apply_1, wherein the processing completion time is later than the queue entering time of the application form apply_1, and the processing starting time is earlier than the queue entering time of the application form apply_1.
The method for calculating the urgent number of the application forms in the corresponding risk level when the application form application_1 enters the waiting queue comprises the following steps: counting the number of the application forms with the same risk level as the application form apply_1, wherein the time of entering the queue in the application form apply_1 is later than the time of entering the queue in the application form apply_1, and the processing starting time is earlier than the processing starting time of the application form apply_1.
After the characteristic information of each application form is acquired, the characteristic information of each automobile financial loan application form can be normalized so as to facilitate subsequent model training, wherein the normalization formula is as follows
Zi is the characteristic parameter to be normalized. And storing the coding parameters (comprising the average coding coefficient and the onehot mapping relation) and the normalization parameters into a database, wherein the data in the sample set are normalized data, and the specific flow is shown in figure 1.
Example five
The present embodiment further improves the model training phase scheme based on the data acquisition and processing phase scheme of the first embodiment and the second to fourth embodiments.
In the model training stage, various machine learning models are used for experiments, including the lightgbm and xgboost based on the tree model, and finally, the neural network model with the best effect is selected.
The neural network is a mathematical model similar to human brain nerve synapses, the embodiment uses tensorsurface to construct a 3-layer fully-connected neural network, a sample set is input into the neural network, parameters of the neural network are continuously adjusted through back propagation to fit a label, a final trained neural network model is obtained and stored, and the label is the actual waiting time of each automobile financial loan application form in the sample set.
The model structure of this embodiment is shown in fig. 3, the number of neurons in each layer is 20, 12, 1, and the amount of data is small, so that a simpler network structure is used, and the activation function rotates the relu function through experiments, and the relu function has an effect 0.6% higher than that of the tanh function.
Each neuron may be represented using the following formula:
each nerve layer can be expressed using the following formula:
Y o*1 ＝relu(w o*i ×X i*1 +b o*1 )
wherein X represents the input of the nerve layer, Y represents the output of the nerve layer, w is the parameter or weight of the nerve network model, b is a bias quantity, w is a matrix of output dimension and input dimension, w and b are independently learned by the nerve network according to data, b is a matrix of output dimension and 1, o is the output dimension, i is the input dimension, and each layer of nerve layer has o+i+o adjustable parameters. The input feature information defined in the above embodiment includes a number-related feature of 3×10 fields, a risk level feature of 10 fields, and a time feature of 3 fields, the input data of the neural network model is a 43-dimensional feature vector of each application form, and the output data is 1 number representing the waiting time. In the neural network model, the input of the first layer of nerve layer is 43 dimensionality, and as the number of neurons of each layer is 20, 12 and 1 respectively, the input of the second layer of nerve layer is 20 dimensionality, the input of the third layer of nerve layer is 12 dimensionality, the input of the fourth layer of nerve layer is 12 dimensionality, and the whole model predicts the waiting time through the adjustment of 1301 (43×20+20+20×12+12+12+12×12+12+12×1+1) parameters.
Example six
The present embodiment improves on the basis of embodiment five, optimizing the model training phase. The neural network is trained by taking the characteristic information of each application form in the training set as the input of a model, and the model outputs a predicted waiting time to be compared with the actual waiting time of the corresponding application form, so that the parameters of the model are continuously adjusted to enable the error to be as small as possible.
In the embodiment, after each training reaches a set number of times (for example, 3 times), calculating an error value of the waiting time predicted according to the current neural network model and the actual waiting time of the corresponding application form as a current error; comparing the current error with a set reference value (for example, setting the reference value as the error is 10 minutes), if the current error is smaller than the reference value, indicating that the error of the current model is acceptable, comparing the current error with a minimum error value obtained in the previous training (namely, a value with the best error effect obtained in the previous training), if the current error is smaller, indicating that the current model is more accurate in prediction, storing the weight of the current neural network model, and continuing training; if the current error is not smaller than the reference value, the error of the current model is too large to be accepted, and the next training is directly carried out; if the current error is not smaller than the minimum error value obtained in the previous training, the prediction accuracy of the current model is not as good as that of the model corresponding to the minimum error value stored previously, and the weight of the current neural network model is not stored at the moment, and the next training is continued.
In practical training, the best model can be obtained only by using a part of sample sets, so the embodiment also provides that when the current error in multiple comparisons is not smaller than the reference value or the minimum error value obtained in the previous training, the training is stopped in advance, and the over-fitting is prevented.
Experiments show that the average absolute error of prediction is 3.8 minutes, and the accuracy of the error within 4 minutes reaches 96.2% on the basis of the waiting time prediction in the automotive loan field by adopting the scheme of the embodiment.
Example seven
As shown in fig. 2, the present embodiment combines the specific solutions of the above embodiments, and proposes a method for training and deploying an automation model, so as to reduce the burden of operators and avoid causing data errors.
At the data acquisition and processing stage, the ELT script is utilized to extract data from a big data platform, automatic data pulling is realized, an application form processing flow table, an application form grade rule table and an approver grade rule table are obtained, simple data processing such as table connection is carried out, and the final production field is about 40 ten thousand of application form numbers, application form grades, queue entering time, processing starting time, processing finishing time, approvers, approver grades, urgent marks and the like.
And then Python realizes data cleaning, performs characteristic engineering, sorts all application forms according to the time of entering the queue, selects the application forms according to the sequence, acquires the accumulated quantity, the checking quantity and the urgent quantity in all application forms corresponding to the risk level of the application forms, extracts the time characteristics of the time of entering the queue, the processing starting time and the processing finishing time of the application forms, increases the risk level and the label (namely the actual waiting time, which is the processing starting time minus the time of entering the queue) of the application forms, repeats the operation until all the application forms are traversed, and generates 43-dimensional characteristics and 1 label for each application form.
And coding the characteristic information (such as mean value coding of time characteristics and onehot coding of risk level characteristics) and normalizing, and then storing the obtained parameters including coding parameters, coding rules (such as onehot coding mapping) and normalization parameters as model deployment data.
In the model training stage, a tensorflow training model is used for outputting a model, model effect evaluation is carried out, a trained neural network model is deployed in a restful mode, and meanwhile, a prediction result is stored in a database.
In the prediction stage, firstly, inputting an application form needing to be predicted for waiting time, recording the currently selected application form as application_2, and automatically acquiring the application form grade, queuing list and application form data currently being audited by auditors from a database by the ELT according to the application form number of the application_2.
And then, carrying out data processing to obtain time features, risk level features and quantity related features corresponding to the application forms needing to be predicted for waiting time, wherein the accumulated quantity in the quantity related features is the quantity of the application forms with the time of entering the queue earlier than the time of the application_2 entering the queue in the queuing list, the checking quantity can be directly obtained, and the urgent quantity is the quantity of the application forms with the urgent marks in the queuing list. After extracting the time feature, the application_2 uses average coding, and at the moment, the rule stored before reading is not needed to be recalculated; the risk level feature of apply_2 is onehot coded, and the rule stored before reading is not needed to be recalculated. All data are normalized, and the rule stored before reading is not needed to be recalculated. And finally generating a 43-dimensional feature corresponding to the application_2 as an input of the model.
And finally, loading the 43-dimensional features corresponding to the obtained apply_2 into a model stored in a model training stage, and automatically predicting the model, returning a prediction result and predicting the number of minutes to wait. The application form number and the prediction result are stored in a database, the waiting time is stored, the training stage can be restarted after a certain number is accumulated, and the iteration is automatically updated. The number of accumulations can be set by itself, such as by setting a quarter auto-update iteration once.
The foregoing is merely a preferred embodiment of the present invention, and it should be noted that modifications and additions may be made to those skilled in the art without departing from the method of the present invention, which modifications and additions are also to be considered as within the scope of the present invention.
Claims (9)
1. A method for estimating the waiting time of an automotive financial loan, which is characterized by comprising the following steps:
data acquisition and processing stage: collecting a plurality of automobile financial loan application forms, extracting and encoding characteristic information of each automobile financial loan application form, wherein the characteristic information comprises time characteristics, risk level characteristics and quantity-related characteristics, and the quantity-related characteristics comprise the accumulation quantity, the checking quantity and the urgent quantity of the application forms in the corresponding risk level of each automobile financial loan application form when entering a waiting queue;
model training stage: constructing a neural network model, and training and verifying the neural network model by utilizing a sample set, wherein the sample set comprises a coded value of characteristic information corresponding to each automobile financial loan application form and actual waiting time, which are obtained in the data acquisition and processing stage; the neural network model takes the coding value of the characteristic information of the automotive financial loan application form as input and takes the actual waiting time as output as a target;
prediction stage: and acquiring an automobile financial loan application form to be predicted, extracting corresponding characteristic information, encoding, inputting the encoded characteristic information into a trained neural network model, and outputting predicted waiting time by the trained neural network model.
2. The method of claim 1, wherein the temporal features extracted during the data acquisition and processing stage are represented by (a, b, c), wherein a, b, and c are positive integers, and a e [1,7] represents monday to sunday, respectively; dividing 24 hours a day into a time period every 30 minutes, b epsilon [0, 47] indicating that b time periods of the day have elapsed; c.epsilon.0, 29 represents the c-th minute in the b+1 time period.
3. The method according to claim 1 or 2, wherein, in the data acquisition and processing stage, the time feature is mean-coded to obtain a coded value of the time feature, the formula is as follows:
wherein x is the time characteristic to be encoded, y is the target value, n i Is x=x i The number of samples in the time, N is the total number of samples, the coefficient
4. The method of claim 1, wherein the risk level features are onehot coded to obtain coded values of the risk level features during the data acquisition and processing stage.
5. The method of claim 1, wherein the temporal features extracted by the data acquisition and processing stage include an auto-finance loan application form in-queue time, a processing start time, and a processing completion time.
6. The method for estimating a waiting time of an automotive finance loan according to claim 5, wherein in the data acquisition and processing stage, the method for calculating the accumulation number of the application forms in the risk level corresponding to any application form when entering the waiting queue is as follows: counting the number of the application forms, of which the processing completion time is later than the queue entering time of the application form apply_1 and the queue entering time is earlier than the queue entering time of the application form apply_1, in the application form which needs to be calculated is apply_1 and has the same risk level as the application form apply_1;
the method for calculating the auditing quantity of the application forms in the corresponding risk level when the application form application_1 enters the waiting queue comprises the following steps: counting the number of the application forms with the same risk level as the application form apply_1, wherein the processing completion time in the application form apply_1 is later than the queue entering time of the application form apply_1, and the processing starting time is earlier than the queue entering time of the application form apply_1;
the method for calculating the urgent number of the application forms in the corresponding risk level when the application form application_1 enters the waiting queue comprises the following steps: counting the number of the application forms with the same risk level as the application form apply_1, wherein the time of entering the queue in the application form apply_1 is later than the time of entering the queue in the application form apply_1, and the processing starting time is earlier than the processing starting time of the application form apply_1.
7. The method of claim 1, wherein,
in the model training stage, after each training set time, calculating an error value of the waiting time predicted according to the current neural network model and the actual waiting time of the corresponding application form as a current error;
comparing the current error with a set reference value, if the current error is smaller than the reference value, comparing the current error with a minimum error value obtained in the previous training, and if the current error is smaller, storing the weight of the current neural network model, and continuing training; if the current error is not smaller than the reference value or is not smaller than the minimum error value obtained in the previous training, the weight of the current neural network model is not saved, and the training is continued;
and terminating the training when the current errors in the multiple comparisons are not smaller than the reference value or are not smaller than the minimum error value obtained in the previous training.
8. An apparatus for implementing a method for predicting a waiting time of an automotive financial loan, comprising:
a memory for storing a computer program;
a processor for implementing the steps of a method of estimating an automotive financial loan wait time as recited in any one of claims 1 to 7 when executing the computer program.
9. A computer readable storage medium, wherein a computer program is stored on the computer readable storage medium, the computer program when executed by a processor implementing the steps of a method for estimating a waiting time of an automotive financial loan according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111220745.5A CN113837873B (en) | 2021-10-20 | 2021-10-20 | Method, equipment and computing medium for estimating waiting time of automobile financial loan |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111220745.5A CN113837873B (en) | 2021-10-20 | 2021-10-20 | Method, equipment and computing medium for estimating waiting time of automobile financial loan |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113837873A CN113837873A (en) | 2021-12-24 |
| CN113837873B true CN113837873B (en) | 2023-05-16 |
Family
ID=78965400
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111220745.5A Active CN113837873B (en) | 2021-10-20 | 2021-10-20 | Method, equipment and computing medium for estimating waiting time of automobile financial loan |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113837873B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104103113A (en) * | 2013-04-15 | 2014-10-15 | 株式会社日立制作所 | Queuing server and queuing method |
| AU2020101475A4 (en) * | 2020-07-24 | 2020-08-27 | Cheng, Fang Miss | A Financial Data Analysis Method Based on Machine Learning Models |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109558989B (en) * | 2018-12-13 | 2023-06-27 | 平安科技（深圳）有限公司 | Queuing time prediction method, queuing time prediction device, queuing time prediction equipment and computer readable storage medium |
| US11521161B2 (en) * | 2019-01-30 | 2022-12-06 | Walmart Apollo, Llc | Automatic determination of pickup wait times |
| CN111489026A (en) * | 2020-04-08 | 2020-08-04 | 中国银行股份有限公司 | Queuing time estimation method, system and device |
-
2021
- 2021-10-20 CN CN202111220745.5A patent/CN113837873B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104103113A (en) * | 2013-04-15 | 2014-10-15 | 株式会社日立制作所 | Queuing server and queuing method |
| AU2020101475A4 (en) * | 2020-07-24 | 2020-08-27 | Cheng, Fang Miss | A Financial Data Analysis Method Based on Machine Learning Models |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113837873A (en) | 2021-12-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111563706A (en) | Multivariable logistics freight volume prediction method based on LSTM network | |
| CN109389494B (en) | Loan fraud detection model training method, loan fraud detection method and device | |
| CN109785618B (en) | Short-term traffic flow prediction method based on combinational logic | |
| CN112819604A (en) | Personal credit evaluation method and system based on fusion neural network feature mining | |
| CN113298230B (en) | Prediction method based on unbalanced data set generated against network | |
| CN109492748B (en) | Method for establishing medium-and-long-term load prediction model of power system based on convolutional neural network | |
| CN110852476A (en) | Passenger flow prediction method and device, computer equipment and storage medium | |
| CN112258251A (en) | Grey correlation-based integrated learning prediction method and system for electric vehicle battery replacement demand | |
| CN112561568A (en) | Target customer prediction method, device and storage medium | |
| CN115470962A (en) | LightGBM-based enterprise confidence loss risk prediction model construction method | |
| Urgun et al. | Composite system reliability analysis using deep learning enhanced by transfer learning | |
| CN115859777A (en) | Method for predicting service life of product system in multiple fault modes | |
| CN112508684B (en) | Collecting-accelerating risk rating method and system based on joint convolutional neural network | |
| CN113837873B (en) | Method, equipment and computing medium for estimating waiting time of automobile financial loan | |
| CN117436653A (en) | Prediction model construction method and prediction method for travel demands of network about vehicles | |
| CN116703568A (en) | Credit card abnormal transaction identification method and device | |
| CN111709844A (en) | Insurance money laundering personnel detection method and device and computer readable storage medium | |
| CN115796030A (en) | Traffic flow prediction method based on graph convolution | |
| CN115641153A (en) | Vehicle price evaluation method based on deep neural network | |
| CN111986018A (en) | Bill collection prompting method and device based on preset collection prompting system and electronic equipment | |
| CN113421154B (en) | Credit risk assessment method and system based on control chart | |
| CN115907181A (en) | Urban rail transit passenger flow prediction method and system | |
| CN114926261A (en) | Method and medium for predicting fraud probability of automobile financial user application | |
| CN115730717A (en) | Power load prediction method and system based on combination of transfer learning strategy and multiple channels | |
| CN115393098A (en) | Financing product information recommendation method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |