WO2020140678A1 - 异常申请检测方法、装置、计算机设备和存储介质 - Google Patents
异常申请检测方法、装置、计算机设备和存储介质 Download PDFInfo
- Publication number
- WO2020140678A1 WO2020140678A1 PCT/CN2019/123190 CN2019123190W WO2020140678A1 WO 2020140678 A1 WO2020140678 A1 WO 2020140678A1 CN 2019123190 W CN2019123190 W CN 2019123190W WO 2020140678 A1 WO2020140678 A1 WO 2020140678A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- application
- feature set
- feature
- preset
- sub
- 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.)
- Ceased
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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
- G06Q40/02—Banking, e.g. interest calculation or account maintenance
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/19—Recognition using electronic means
- G06V30/192—Recognition using electronic means using simultaneous comparisons or correlations of the image signals with a plurality of references
- G06V30/194—References adjustable by an adaptive method, e.g. learning
Definitions
- This application relates to an abnormal application detection method, device, computer equipment, and storage medium.
- an abnormal application detection method, device, computer equipment, and storage medium are provided.
- An abnormal application detection method including:
- the model is constructed based on the unmarked training feature set, which is used to perform anomaly feature detection by applying the features in the feature set to the corresponding path length in the isolated tree of the isolated forest model, and based on the length of the path for the application feature set Score;
- An abnormal application detection device including:
- the data acquisition module is used to obtain the credit application data of the applicant
- the feature extraction module is used to obtain the applicant information and the application feature set corresponding to the applicant information according to the credit application data, and the application features in the application feature set include application time, application frequency, application trend and institution preference ;
- a score calculation module configured to input the application feature set into a preset isolated forest model, perform anomaly detection on the features in the application feature set through the preset isolated forest model, and obtain an application score corresponding to the application feature set
- the preset isolated forest model is constructed based on the unmarked training feature set, which is used to perform abnormal feature detection by applying the features in the feature set to the corresponding path length in the isolated tree of the isolated forest model, and based on the path Length to score the application feature set;
- the abnormal application determination module is used to obtain the absolute value of the difference between the application score and the preset reference value, and when the absolute value of the difference is lower than the preset threshold, the application corresponding to the credit application data is determined to be an abnormal application .
- a computer device includes a memory and one or more processors.
- the memory stores computer-readable instructions.
- the one or more processors are executed The following steps:
- the model is constructed based on the unmarked training feature set, which is used to perform anomaly feature detection by applying the features in the feature set to the corresponding path length in the isolated tree of the isolated forest model, and based on the length of the path for the application feature set Score;
- One or more non-volatile storage media storing computer readable instructions.
- the computer readable instructions When executed by one or more processors, the one or more processors perform the following steps:
- the model is constructed based on the unmarked training feature set, which is used to perform anomaly feature detection by applying the features in the feature set to the corresponding path length in the isolated tree of the isolated forest model, and based on the length of the path for the application feature set Score;
- FIG. 1 is an application scenario diagram of an abnormal application detection method according to one or more embodiments.
- FIG. 2 is a schematic flowchart of an abnormal application detection method according to one or more embodiments.
- FIG. 3 is a schematic flowchart of sub-steps of step S400 in FIG. 2 according to one or more embodiments.
- FIG. 4 is a schematic flowchart of an abnormal application detection method in another embodiment.
- FIG. 5 is a schematic flowchart of sub-steps of step S600 in FIG. 2 according to one or more embodiments.
- FIG. 6 is a block diagram of an abnormal application detection device according to one or more embodiments.
- FIG. 7 is a block diagram of a computer device according to one or more embodiments.
- the abnormal application detection method provided by this application can be applied to the application environment shown in FIG. 1.
- the terminal 102 and the server 104 communicate via the network.
- the server 104 obtains the credit application data of the applicant provided by the application monitoring staff through the network. After receiving the credit application data, the server first obtains the applicant information and the application feature set corresponding to the applicant information based on the credit application data, and then applies the application characteristics Input the preset isolated forest model, use the preset isolated forest model to perform anomaly detection on the features in the application feature set, obtain the application score corresponding to the application feature set, and then determine whether the application is an abnormal application according to the application score, and then detect the The result is fed back to the terminal 102.
- the terminal 102 may be, but not limited to, various personal computers, notebook computers, smart phones, tablets, and portable wearable devices.
- the server 104 may be implemented by an independent server or a server cluster composed of multiple servers.
- an abnormal application detection method is provided.
- the method is applied to the server in FIG. 1 as an example for illustration, including the following steps:
- Applicant specifically refers to the applicant who is applying for a credit.
- the application specifically refers to a credit application.
- the credit application data specifically includes the data provided by the user for the credit application, which can specifically include the user's personal information and the credit that the user can provide. data.
- Applicant information refers to the personal information corresponding to the applicant. According to the applicant information, the relevant data of the applicant can be obtained.
- the application feature set specifically refers to the feature data used to analyze the applicant's credit situation.
- the application feature set can specifically include the application Time dimension, frequency of application, application trend, institutional preference and other dimensions.
- the isolated forest model is a model based on the isolated forest algorithm.
- the isolated forest algorithm is generally used to mine abnormal data, or outlier mining, that is, in a large number of data, find the data that does not conform to the rules of other data. . For the anomaly data found, then either the anomaly data should be directly cleared, such as the noise removal data in the data cleaning, or the in-depth analysis of the anomaly data, such as the analysis of the behavioral characteristics of attacks and fraud.
- the preset isolated forest model is based on the unmarked training feature set and is used to detect whether the application belongs to an abnormal application model. The preset isolated forest model is used to detect the applicant's previous application characteristics to obtain the application corresponding to the application data. score.
- the application scoring process includes obtaining the length of the corresponding path of each feature in the application feature set in the isolated tree of the isolated forest model, performing abnormal feature detection based on the path length, and scoring the application feature set based on the path length to obtain the current Apply for the corresponding score.
- the application score refers to the score obtained based on the application data.
- the application feature set is input into the preset isolated forest model to obtain the application score corresponding to the application feature set.
- each application feature in the application feature set is anomaly detected through an isolated tree in the preset forest model.
- An application feature can be detected by multiple isolated trees to obtain anomalies corresponding to each application feature Degree, and then integrate the abnormality of each application feature to obtain the application score corresponding to the application feature set.
- the abnormal application specifically includes a fraud application, that is, the applicant's credit application contains fraud.
- a fraud application that is, the applicant's credit application contains fraud.
- anomalies anomaly detection
- anomalies are defined as "more easily like outliers (more likely like to be separated)", which can be understood as sparse and isolated Groups with higher density are farther away.
- sparsely distributed areas indicate that the probability of events occurring in this area is very low, so it can be considered that the data falling in these areas is abnormal.
- the application corresponding to the credit application data is determined to be an abnormal application.
- the preset benchmark value is 1, and the closer the application score is to 1, the higher the probability of fraud.
- an abnormal threshold may be set, and then the difference between the application score and 1 is obtained.
- the absolute value of the difference between the application score and 1 is lower than or equal to the abnormal threshold, the application is determined to be an abnormal application. If the difference from 1 is higher than the abnormal threshold, it is determined that the application is not an abnormal application. .
- the above abnormal application detection method first obtains the applicant's credit application data; obtains the applicant information and the application feature set corresponding to the applicant information based on the credit application data; enters the application feature set into the preset isolated forest model after unsupervised training, Anomaly detection is performed on the features in the application feature set through a preset isolated forest model to obtain the application score corresponding to the application feature set; according to the application score, it is determined whether the application corresponding to the credit application data is an abnormal application.
- This proposal uses the default isolated forest model obtained by unsupervised learning to score credit applications without labeling training, which greatly improves the practicality of the abnormal application scoring system and has the ability to identify variant frauds and frauds that have never been seen before.
- the output in the form of score reflects its abnormal degree, which is convenient for customers to understand and productize.
- S400 specifically includes:
- before S600 includes:
- the unmarked training feature set refers to the feature set used to train the isolated forest model.
- the composition of the training feature set is similar to that of the application feature set, and includes dimensions such as application time, application frequency, application trend, and institutional preference.
- the isolated forest model is composed of multiple isolated trees. First, an isolated tree is established based on the unmarked training feature set, and an isolated forest model used for application scoring according to the application feature set is composed of multiple isolated trees.
- the unlabeled training feature set includes features such as application time, application frequency, application trends, and institutional preferences.
- the data can build an isolated tree, and then train to generate multiple isolated trees based on a large amount of unlabeled data to form an isolated forest.
- S520 specifically includes: sampling a sample feature set containing ⁇ samples from the unlabeled training feature set as an isolated tree training sample set; randomly selecting individual features of the samples in the sample feature set; according to the features Binary divide the sample feature set to obtain two sub-feature sets; determine whether the sub-feature set can be divided again, and when the sub-feature set can be divided again, return the sub-feature set as a new sample feature set The step of performing binary division to obtain two sub-feature sets; when the sub-feature set cannot be divided again, the binary division of the sub-feature set is stopped.
- the process of constructing an isolated tree based on the unlabeled training feature set specifically includes: first, a sample feature set containing ⁇ samples is sampled from the training feature set containing the training sample as the training sample set of the isolated tree, and based on the sample feature set, a Isolated tree for training.
- each isolated tree may be equally divided into a training feature set based on the number of isolated trees, and the isolated tree may be trained based on the divided training feature set.
- the sample feature set After establishing the sample feature set, randomly select a feature contained in the sample in the sample feature set, and obtain the possible value of this feature, and then randomly select a value in all the value ranges of this feature, and perform a binary search on the sample feature set Divide, divide the samples in the sample feature set less than this value to the left of the node, and the samples greater than or equal to this value to the right of the node. This results in a split condition and sub-feature sets on both sides. At the same time, it is judged whether the sub-feature set can be divided again. When the sub-feature can be divided again, a value is randomly selected from all the value ranges of the feature in this sub-feature set, and the sub-feature set is divided into two parts.
- the binary division of the sub-feature set ends.
- the division referred to here is true for all sub-feature sets that can be divided.
- the isolation tree can be effectively established by dividing the feature set. And can effectively detect abnormal applications based on the feature tree.
- the sample feature set is binary-divided according to the features, and after obtaining the two sub-feature sets includes:
- the number of returns needs to be recorded.
- the return number is greater than log2( ⁇ )-2
- the binary division of the sub-feature set needs to be stopped.
- the height of the isolated tree is limited. The height of the isolated tree is determined based on the number of samples ⁇ contained in the sample feature set used for training. When the height of the isolated tree reaches log2( ⁇ ). The height of the isolated tree and whether the sample can be divided again is the basis for whether the sub-feature set needs to be divided again. If either side meets the corresponding conditions, the binary division should be terminated.
- step S600 specifically includes:
- S660 Determine the application score of the application feature set according to the path length corresponding to all the application features in the application feature set.
- the abnormal application detection method of the present application specifically includes the following steps: obtaining the applicant's credit application data; obtaining the applicant information based on the credit application data; based on the applicant information, looking up the applicant's Historical application data to obtain historical application data; based on historical application data to obtain application feature set, the application features in the application feature set include application time, application frequency, application trends and institutional preferences; sampling from the unmarked training feature set contains ⁇
- the sample feature set of the sample is used as the training sample set of the isolated tree; the single feature of the sample in the sample feature set is randomly selected; the sample feature set is binary divided according to the feature to obtain two sub-feature sets; whether the sub-feature set can be divided again , When the sub-feature set can be divided again, return the sub-feature set as a new sample feature set to binary divide the sample feature set according to the feature, obtain two sub-feature sets, and record the number of returns; when the sub-feature set cannot When dividing again or when the number of returns is greater than log2( ⁇ )-2, the binary
- steps in the flowcharts of FIGS. 2-5 are sequentially displayed in accordance with the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless clearly stated in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least some of the steps in FIGS. 2-5 may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but may be executed at different times. These sub-steps or stages The execution order of is not necessarily sequential, but may be executed in turn or alternately with at least a part of other steps or sub-steps or stages of other steps.
- an abnormality application detection device including: a data acquisition module 200, a feature extraction module 400, a score calculation module 600, and an abnormality application determination module 800, wherein:
- the data obtaining module 200 is used to obtain the credit application data of the applicant
- the feature extraction module 400 is used to obtain the applicant information and the application feature set corresponding to the applicant information according to the credit application data;
- the score calculation module 600 is used to input the application feature set into the preset isolated forest model, perform anomaly detection on the features in the application feature set through the preset isolated forest model, and obtain the application score corresponding to the application feature set.
- the preset isolated forest model is based on no
- the labeled training feature set is constructed to perform anomaly feature detection by applying the features in the feature set to the corresponding path length in the isolated tree of the isolated forest model, and scoring the applied feature set based on the length of the path;
- the abnormal application determination module 800 is used to obtain the absolute value of the difference between the application score and the preset reference value. When the absolute value of the difference is lower than the preset threshold, the application corresponding to the credit application data is determined to be an abnormal application.
- the feature extraction module 400 is specifically configured to: obtain applicant information based on credit application data; obtain applicant historical application data based on applicant information; and obtain application feature sets based on historical application data.
- the score calculation module 600 is used to: input the application feature set into the preset isolated forest model; perform anomaly detection on each application feature in the application feature set through the isolated tree in the preset forest model to obtain the respective Obtain the anomaly degree corresponding to each application feature; and obtain the application score corresponding to the application feature set according to the anomaly degree of each application feature.
- the model training module includes:
- the isolated tree building unit is used to construct an isolated tree by preset unmarked training feature sets;
- the model building unit constructs a preset isolated forest model based on the isolated tree.
- the isolated tree building unit is specifically used for: sampling a sample feature set containing ⁇ samples from the unlabeled training feature set as an isolated tree training sample set; randomly selecting a single sample in the sample feature set Features; Binary divide the sample feature set according to the feature to obtain two sub-feature sets; determine whether the sub-feature set can be divided again, and when the sub-feature set can be divided again, return the sub-feature set as a new sample feature set according to the feature. The step of bifurcation of the sample feature set to obtain two sub-feature sets; and when the number of samples in the sub-feature set is equal to 1, stop the bifurcation of the sub-feature set.
- the isolated tree building unit is also used to: return the sub-feature set as a new sample feature set.
- the score calculation module 600 is specifically used to: obtain the application features in the application feature set; run the corresponding application features on the isolated tree in the preset isolated forest model, and record the path that the application features traverse during the operation Length; and determine the application score of the application feature set according to the path length corresponding to all the application features in the application feature set.
- the abnormal application determination module 800 is specifically configured to: obtain the difference between the application score and 1, and when the difference is lower than a preset threshold, determine that the application is an abnormal application.
- Each module in the above abnormal application detection device may be implemented in whole or in part by software, hardware, or a combination thereof.
- the above modules may be embedded in the hardware form or independent of the processor in the computer device, or may be stored in the memory in the computer device in the form of software so that the processor can call and execute the operations corresponding to the above modules.
- a computer device is provided.
- the computer device may be a server, and its internal structure may be as shown in FIG. 7.
- the computer device includes a processor, memory, network interface, and database connected by a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities.
- the memory of the computer device includes a non-volatile storage medium and an internal memory.
- the non-volatile storage medium stores an operating system, a computer program, and a database.
- the internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium.
- the database of the computer device is used to store abnormal application data.
- the network interface of the computer device is used to communicate with external terminals through a network connection. When the computer program is executed by the processor, an abnormal application detection method is realized.
- FIG. 7 is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied.
- the specific computer device may Include more or less components than shown in the figure, or combine certain components, or have a different arrangement of components.
- a computer device includes a memory and one or more processors.
- the memory stores computer-readable instructions.
- the steps of the abnormal application detection method provided in any embodiment of the present application are implemented.
- One or more non-volatile storage media storing computer-readable instructions, which when executed by one or more processors, cause the one or more processors to implement the exception provided in any embodiment of the present application Apply the steps of the detection method.
- Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.
- Volatile memory can include random access memory (RAM) or external cache memory.
- RAM random access memory
- DRAM dynamic RAM
- SDRAM synchronous DRAM
- DDRSDRAM double data rate SDRAM
- ESDRAM enhanced SDRAM
- SLDRAM synchronous chain (Synchlink) DRAM
- RDRAM direct RAM
- DRAM direct memory bus dynamic RAM
- RDRAM memory bus dynamic RAM
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Accounting & Taxation (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Databases & Information Systems (AREA)
- Finance (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- Technology Law (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
一种异常申请检测方法,包括:获取申请人的信贷申请数据;根据所述信贷申请数据获取申请人信息以及与所述申请人信息对应的申请特征集;将所述申请特征集输入预设孤立森林模型,通过所述预设孤立森林模型对所述申请特征集内特征进行异常检测,获取所述申请特征集对应的申请评分,所述预设孤立森林模型基于无标记的训练特征集构建;及获取所述申请评分与预设基准值差值的绝对值,当所述差值的绝对值低于预设阈值时,判定所述信贷申请数据对应的申请为异常申请。
Description
相关申请的交叉引用
本申请要求于2019年01月04日提交中国专利局,申请号为2019100070454,申请名称为“异常申请检测方法、装置、计算机设备以及存储介质”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
本申请涉及一种异常申请检测方法、装置、计算机设备和存储介质。
随着互联网信贷产业在近几年的飞速发展,其呈现出百家争鸣、百花齐放的局面,然而伴随着产业的欣荣发展,欺诈黑色产业链也在不断地渗透到该领域,各种新颖的欺诈模式层出不穷,对互联网信贷产业的健康发展蒙上了一层阴影。据不完全统计,每年因欺诈导致的损失巨大,欺诈风险已成为互联网信贷产业风险的重中之重。
然而,发明人意识到,目前业内主要采取监督学习算法对用户的贷款申请进行欺诈检测,但大部分情况下的用于检测用户行为是否属于欺诈行为的数据是没有标签的,人工进行标注的成本极大,且诈骗的手段总在变化。
发明内容
根据本申请公开的各种实施例,提供一种异常申请检测方法、装置、计算机设备和存储介质。
一种异常申请检测方法,包括:
获取申请人的信贷申请数据;
根据所述信贷申请数据获取申请人信息以及与所述申请人信息对应的申请特征集;
将所述申请特征集输入预设孤立森林模型,通过所述预设孤立森林模型对所述申请特征集内特征进行异常检测,获取所述申请特征集对应的申请评分,所述预设孤立森林模型基于无标记的训练特征集构建,用于通过申请特征集内各个特征在所述孤立森林模型的孤立树中对应路径的长度进行异常特征检测,并基于所述路径的长度对申请特征集进行评分;及
获取所述申请评分与预设基准值差值的绝对值,当所述差值的绝对值低于预设阈值时,判定所述信贷申请数据对应的申请为异常申请。
一种异常申请检测装置,包括:
数据获取模块,用于获取申请人的信贷申请数据;
特征提取模块,用于根据所述信贷申请数据获取申请人信息以及与所述申请人信息对应的申请特征集,所述申请特征集中的申请特征包括申请时点、申请频次、申请趋势以及机构偏好;
评分计算模块,用于将所述申请特征集输入预设孤立森林模型,通过所述预设孤立森林模型对所述申请特征集内特征进行异常检测,获取所述申请特征集对应的申请评分,所述预设孤立森林模型基于无标记的训练特征集构建,用于通过申请特征集内各个特征在所述孤立森林模型的孤立树中对应路径的长度进行异常特征检测,并基于所述路径的长度对申请特征集进行评分;及
异常申请判定模块,用于获取所述申请评分与预设基准值差值的绝对值,当所述差值的绝对值低于预设阈值时,判定所述信贷申请数据对应的申请为异常申请。
一种计算机设备,包括存储器和一个或多个处理器,所述存储器中储存有计算机可读指令,所述计算机可读指令被所述处理器执行时,使得所述一个或多个处理器执行以下步骤:
获取申请人的信贷申请数据;
根据所述信贷申请数据获取申请人信息以及与所述申请人信息对应的申请特征集;
将所述申请特征集输入预设孤立森林模型,通过所述预设孤立森林模型对所述申请特征集内特征进行异常检测,获取所述申请特征集对应的申请评分,所述预设孤立森林模型基于无标记的训练特征集构建,用于通过申请特征集内各个特征在所述孤立森林模型的孤立树中对应路径的长度进行异常特征检测,并基于所述路径的长度对申请特征集进行评分;及
获取所述申请评分与预设基准值差值的绝对值,当所述差值的绝对值低于预设阈值时,判定所述信贷申请数据对应的申请为异常申请。
一个或多个存储有计算机可读指令的非易失性存储介质,计算机可读指令被一个或多个处理器执行时,使得一个或多个处理器执行以下步骤:
获取申请人的信贷申请数据;
根据所述信贷申请数据获取申请人信息以及与所述申请人信息对应的申请特征集;
将所述申请特征集输入预设孤立森林模型,通过所述预设孤立森林模型对所述申请特征集内特征进行异常检测,获取所述申请特征集对应的申请评分,所述预设孤立森林模型基于无标记的训练特征集构建,用于通过申请特 征集内各个特征在所述孤立森林模型的孤立树中对应路径的长度进行异常特征检测,并基于所述路径的长度对申请特征集进行评分;及
获取所述申请评分与预设基准值差值的绝对值,当所述差值的绝对值低于预设阈值时,判定所述信贷申请数据对应的申请为异常申请。
本申请的一个或多个实施例的细节在下面的附图和描述中提出。本申请的其它特征和优点将从说明书、附图以及权利要求书变得明显。
为了更清楚地说明本申请实施例中的技术方案,下面将对实施例中所需要使用的附图作简单地介绍,显而易见地,下面描述中的附图仅仅是本申请的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动的前提下,还可以根据这些附图获得其它的附图。
图1为根据一个或多个实施例中异常申请检测方法的应用场景图。
图2为根据一个或多个实施例中异常申请检测方法的流程示意图。
图3为根据一个或多个实施例中图2中步骤S400的子步骤的流程示意图。
图4为另一个实施例中异常申请检测方法的流程示意图。
图5为根据一个或多个实施例中图2中步骤S600的子步骤的流程示意图。
图6为根据一个或多个实施例中异常申请检测装置的框图。
图7为根据一个或多个实施例中计算机设备的框图。
为了使本申请的技术方案及优点更加清楚明白,以下结合附图及实施例,对本申请进行进一步详细说明。应当理解,此处描述的具体实施例仅仅用以解释本申请,并不用于限定本申请。
本申请提供的异常申请检测方法,可以应用于如图1所示的应用环境中。终端102与服务器104通过网络进行通信。服务器104通过网络获取申请监测工作人员提供的申请人的信贷申请数据,服务器接受到信贷申请数据后,首先根据信贷申请数据获取申请人信息以及与申请人信息对应的申请特征集,而后将申请特征集输入预设孤立森林模型,通过预设孤立森林模型对申请特征集内特征进行异常检测,获取申请特征集对应的申请评分,而后根据该申请评分判断该申请是否属于异常申请,而后将检测的结果反馈至终端102。终端102可以但不限于是各种个人计算机、笔记本电脑、智能手机、平板电脑和便携式可穿戴设备,服务器104可以用独立的服务器或者是多个服 务器组成的服务器集群来实现。
在其中一个实施例中,如图2所示,提供了一种异常申请检测方法,以该方法应用于图1中的服务器为例进行说明,包括以下步骤:
S200,获取申请人的信贷申请数据。
申请人具体是指正在进行信贷申请的申请人,申请具体指的是信贷申请,信贷申请数据具体包括用户用于信贷申请所提供的数据,具体可以包括用户的个人信息,以及用户能提供的信贷数据。
在申请人提出信贷申请时,如果需要判断该申请是否处于异常申请,首先需要获得申请人用于申请提供的信贷申请数据。
S400,根据信贷申请数据获取申请人信息以及与申请人信息对应的申请特征集;
申请人信息指的是与申请人对应的个人信息,根据申请人信息可以获取申请人相关的数据,申请特征集具体是指用于分析申请人信贷情况的特征数据,申请特征集具体可以包括申请时点、申请频次、申请趋势以及机构偏好等等维度。
根据用户提交的信贷申请数据确定申请人,并获取申请人信息,同时获取申请人在信贷申请方面的申请特征集。
S600,将申请特征集输入预设孤立森林模型,通过预设孤立森林模型对申请特征集内特征进行异常检测,获取申请特征集对应的申请评分,用于通过申请特征集内各个特征在孤立森林模型的孤立树中对应路径的长度进行异常特征检测,并基于路径的长度对申请特征集进行评分。
孤立森林模型是基于孤立森林算法建立的模型,孤立森林算法一般用于挖掘异常数据,或者说离群点挖掘,即是在一大堆数据中,找出与其它数据的规律不太符合的数据。对于找出的异常数据,然后要么直接清除异常数据,如数据清理中的去除噪声数据,要么深入分析异常数据,比如分析攻击、欺诈的行为特征。而预设孤立森林模型基于无标记的训练特征集构建,用于检测申请是否属于异常申请的模型,通过预设孤立森林模型对申请人过往的申请特征进行检测,获得该次申请数据对应的申请评分。获取申请评分过程包括,获取申请特征集内各个特征在孤立森林模型的孤立树中对应路径的长度,基于路径长度进行异常特征检测,并基于路径的长度对申请特征集进行评分,来获得当前的申请对应的评分。申请评分是指基于该申请数据得到的评分。
在获得包含申请特征的申请特征集后,将申请特征集输入预设的孤立森林模型,获取申请特征集对应的申请评分。在其中一个实施例中,通过预设森林模型中的孤立树对申请特征集内的各个申请特征进行异常检测,一个申请特征可以通过多个孤立树进行检测,分别得到每个申请特征对应的异常度, 而后综合各个申请特征的异常度,得到该申请特征集对应的申请评分。
S800,获取申请评分与预设基准值差值的绝对值,当差值的绝对值低于预设阈值时,判定信贷申请数据对应的申请为异常申请。
异常申请具体包括欺诈申请,即申请人的该信贷申请包含欺诈行为。目前学术界对异常(anomaly detection)的定义有很多种,在孤立森林中,异常被定义为“容易被孤立的离群点(more likely to be separated)”,可以将其理解为分布稀疏且离密度高的群体较远的点。在特征空间里,分布稀疏的区域表示事件发生在该区域的概率很低,因而可以认为落在这些区域里的数据是异常的。
在得到基于预设孤立森林模型得到的申请评分后,可以基于该申请评分对该申请是否属于异常申请进行检测,获取申请评分与预设基准值差值的绝对值,当差值的绝对值低于预设阈值时,判定信贷申请数据对应的申请为异常申请。一般来说,预设基准值为1,该申请评分越接近1,其存在欺诈的可能性越高。在其中一个实施例中,可以设置一个异常阈值,而后获取申请评分与1的差值,当申请评分与1的差的绝对值低于或等于异常阈值时,判定该申请属于异常申请,当评分与1的差的高于异常阈值是,判定该申请不属于异常申请。。
上述异常申请检测方法,首先获取申请人的信贷申请数据;根据信贷申请数据获取申请人信息以及与申请人信息对应的申请特征集;将申请特征集输入经过无监督训练获得预设孤立森林模型,通过预设孤立森林模型对申请特征集内特征进行异常检测,获取申请特征集对应的申请评分;根据申请评分判断信贷申请数据对应的申请是否属于异常申请。本提案采用无监督学习获得的预设孤立森林模型对信贷申请进行评分,无需标签进行训练,大大提高了异常申请评分系统的实用性,有能力识别变种的诈骗和从未见过的诈骗,以评分的形式输出,反映其异常程度,便于客户理解与产品化。
如图3所示,在其中一个实施例中,S400具体包括:
S420,根据信贷申请数据获取申请人信息。
S440,根据申请人信息,获取申请人的历史申请数据。
S460,根据历史申请数据获得申请特征集。
首先基于申请人提出的申请确定申请人的信息,基于该申请人的信息在信贷数据库内进行搜索,查找申请人的历史信贷申请记录,并获取这些历史申请数据。并基于这些历史申请记录对申请人的申请特征进行总结,获得申请特征集,通过对申请人历史申请记录的总结,可以有效获得申请人的申请特征,并基于申请人的申请特征对申请人的申请是否属于异常申请进行检测。
如图4所示,在其中一个实施例中,S600之前包括:
S520,通过预设无标记的训练特征集构建孤立树;
S540,根据孤立树构建预设的孤立森林模型。
无标记的训练特征集是指用于训练孤立森林模型的特征集,训练特征集的组成与申请特征集类似,都包括申请时点、申请频次、申请趋势以及机构偏好等维度。孤立森林模型由多个孤立树组成,首先基于无标记的训练特征集建立孤立树,通过多个孤立树组成用于根据申请特征集进行申请评分的孤立森林模型。无标记的训练特征集包括申请时点、申请频次、申请趋势以及机构偏好等特征,可以随机选取无标记的训练特征集中的某个特征建立一个孤立树,如基于训练特征集中申请人的申请频次的数据可以建立一个孤立树,而后基于大量的无标记数据训练生成多个孤立树组成孤立森林。
在其中一个实施例中,S520具体包括:从无标记的训练特征集中抽样出包含ψ个样本的样本特征集,作为孤立树的训练样本集;随机选取样本特征集内样本的单个特征;根据特征对样本特征集进行二叉划分,获得两个子特征集;判断子特征集是否能再次划分,当子特征集能再次划分时,将子特征集作为新的样本特征集返回根据特征对样本特征集进行二叉划分,获得两个子特征集的步骤;当子特征集不能再次划分时,停止对子特征集的二叉划分。
根据无标记的训练特征集构建孤立树的过程具体包括:首先从包含训练样本的训练特征集中抽样出包含ψ个样本的样本特征集作为孤立树的训练样本集,并基于该样本特征集对一个孤立树进行训练。在其中一个实施例中,可以基于孤立树的数量对每个孤立树均分训练特征集,并基于均分后的训练特征集对孤立树进行训练。在确立样本特征集后,随机选取样本特征集内样本所包含的一个特征,并获取这个特征可能的取值,而后在这个特征的所有值范围内随机选一个值,对样本特征集进行二叉划分,将样本特征集中小于该值的样本划分到节点的左边,大于等于该值的样本划分到节点的右边。这样得到了一个分裂条件和两边的子特征集。同时判断子特征集是否可以再次划分,当子特征可以再次划分时,再次在这个子特征集内该特征的所有值范围内随机选一个值,对子特征集进行二叉划分。当不能再次划分时即子特征只包含一个数据时,结束对子特征集的二叉划分,这里指的划分对所有可划分的子特征集都成立。通过对特征集的划分能有效得建立孤立树。并能基于特征树对异常的申请进行有效检测。
在其中一个实施例中,根据特征对样本特征集进行二叉划分,获得两个子特征集之后包括:
将子特征集作为新的样本特征集返回根据特征对样本特征集进行二叉划分,获得两个子特征集的步骤,记录返回次数,当返回次数大于log2(ψ)-2时,停止对子特征集的二叉划分。
在将子特征集作为新的样本特征集进行二叉划分时,需要记录返回的次数,当返回次数大于log2(ψ)-2时,需要停止对子特征集的二叉划分。孤立树的高度有一定限制,孤立树的高度基于训练所用的样本特征集所包含的样本数目ψ确定,当孤立树的高度最高达到log2(ψ)。孤立树的高度与样本是否可以再次划分是是否需要对子特征集再次进行二叉划分的依据,任意一方达到对应条件,都应终止二叉划分。
如图5所示,在其中一个实施例中,步骤S600具体包括:
S620,获取申请特征集中申请特征。
S640,将对应的申请特征在预设孤立森林模型中的孤立树上运行,记录申请特征在运行过程中经过的路径长度。
S660,根据申请特征集内所有申请特征对应的路径长度确定申请特征集的申请评分。
在获得申请人的申请特征集后,提取申请特征集内的申请特征,并将申请特征在对应的孤立树上运行,记录单个申请特征在孤立树上到达不可再次划分的节点所经过的路径长度,通过所有特征的路径长度确定申请特征集对应的申请评分。通过申请特征集内所有特征在孤立森林模型内所跑的路径的长度可以确定该申请的某个特征异常,并可以基于每个特征的长度综合对其进行评分。
在其中一个实施例中,本申请的异常申请检测方法具体包括以下步骤:获取申请人的信贷申请数据;根据信贷申请数据获取申请人信息;根据申请人信息,查找申请人在各类信贷机构的历史申请数据,获取历史申请数据;根据历史申请数据获得申请特征集,申请特征集中的申请特征包括申请时点、申请频次、申请趋势以及机构偏好;从无标记的训练特征集中抽样出包含ψ个样本的样本特征集,作为孤立树的训练样本集;随机选取样本特征集内样本的单个特征;根据特征对样本特征集进行二叉划分,获得两个子特征集;判断子特征集是否能再次划分,当子特征集能再次划分时,将子特征集作为新的样本特征集返回根据特征对样本特征集进行二叉划分,获得两个子特征集的步骤,并记录返回次数;当子特征集不能再次划分时或者当返回次数大于log2(ψ)-2时,停止对子特征集的二叉划分。根据孤立树构建预设的孤立森林模型。获取申请特征集中申请特征;将对应的申请特征在预设孤立森林模型中的孤立树上运行,记录申请特征在运行过程中经过的路径长度;根据申请特征集内所有申请特征对应的路径长度确定申请特征集的申请评分。获取申请评分与1的差值,当差值低于预设阈值时,判定信贷申请数据对应的申请为异常申请。
应该理解的是,虽然图2-5的流程图中的各个步骤按照箭头的指示依次 显示,但是这些步骤并不是必然按照箭头指示的顺序依次执行。除非本文中有明确的说明,这些步骤的执行并没有严格的顺序限制,这些步骤可以以其它的顺序执行。而且,图2-5中的至少一部分步骤可以包括多个子步骤或者多个阶段,这些子步骤或者阶段并不必然是在同一时刻执行完成,而是可以在不同的时刻执行,这些子步骤或者阶段的执行顺序也不必然是依次进行,而是可以与其它步骤或者其它步骤的子步骤或者阶段的至少一部分轮流或者交替地执行。
在其中一个实施例中,如图6所示,提供了一种异常申请检测装置,包括:数据获取模块200、特征提取模块400、评分计算模块600和异常申请判定模块800,其中:
数据获取模块200,用于获取申请人的信贷申请数据;
特征提取模块400,用于根据信贷申请数据获取申请人信息以及与申请人信息对应的申请特征集;
评分计算模块600,用于将申请特征集输入预设孤立森林模型,通过预设孤立森林模型对申请特征集内特征进行异常检测,获取申请特征集对应的申请评分,预设孤立森林模型基于无标记的训练特征集构建,用于通过申请特征集内各个特征在孤立森林模型的孤立树中对应路径的长度进行异常特征检测,并基于路径的长度对申请特征集进行评分;及
异常申请判定模块800,用于获取申请评分与预设基准值差值的绝对值,当差值的绝对值低于预设阈值时,判定信贷申请数据对应的申请为异常申请。
在其中一个实施例中,特征提取模块400具体用于:根据信贷申请数据获取申请人信息;根据申请人信息,获取申请人的历史申请数据;及根据历史申请数据获得申请特征集。
在其中一个实施例中,评分计算模块600用于:将申请特征集输入预设孤立森林模型;通过预设森林模型中的孤立树对申请特征集内的各个申请特征进行异常检测,获取各分别得到每个申请特征对应的异常度;及根据各申请特征的异常度,获取申请特征集对应的申请评分。
在其中一个实施例中,还包括模型训练模块,模型训练模块包括:
孤立树建立单元,用于通过预设无标记的训练特征集构建孤立树;
模型建立单元,根据孤立树构建预设的孤立森林模型。
在其中一个实施例中,孤立树建立单元具体用于:从无标记的训练特征集中抽样出包含ψ个样本的样本特征集,作为孤立树的训练样本集;随机选取样本特征集内样本的单个特征;根据特征对样本特征集进行二叉划分,获得两个子特征集;判断子特征集是否能再次划分,当子特征集能再次划分时,将子特征集作为新的样本特征集返回根据特征对样本特征集进行二叉划分, 获得两个子特征集的步骤;及当子特征集内样本数等于1时,停止对子特征集的二叉划分。
在其中一个实施例中,孤立树建立单元还用于:将子特征集作为新的样本特征集返回根据特征对样本特征集进行二叉划分,获得两个子特征集的步骤,记录返回次数,当返回次数大于log2(ψ)-2时,停止对子特征集的二叉划分。
在其中一个实施例中,评分计算模块600具体用于:获取申请特征集中申请特征;将对应的申请特征在预设孤立森林模型中的孤立树上运行,记录申请特征在运行过程中经过的路径长度;及根据申请特征集内所有申请特征对应的路径长度确定申请特征集的申请评分。
在其中一个实施例中,异常申请判定模块800具体用于:获取申请评分与1的差值,当差值低于预设阈值时,判定申请为异常申请。
关于异常申请检测装置的具体限定可以参见上文中对于异常申请检测方法的限定,在此不再赘述。上述异常申请检测装置中的各个模块可全部或部分通过软件、硬件及其组合来实现。上述各模块可以硬件形式内嵌于或独立于计算机设备中的处理器中,也可以以软件形式存储于计算机设备中的存储器中,以便于处理器调用执行以上各个模块对应的操作。
在其中一个实施例中,提供了一种计算机设备,该计算机设备可以是服务器,其内部结构图可以如图7所示。该计算机设备包括通过系统总线连接的处理器、存储器、网络接口和数据库。其中,该计算机设备的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性存储介质、内存储器。该非易失性存储介质存储有操作系统、计算机程序和数据库。该内存储器为非易失性存储介质中的操作系统和计算机程序的运行提供环境。该计算机设备的数据库用于存储异常申请数据。该计算机设备的网络接口用于与外部的终端通过网络连接通信。该计算机程序被处理器执行时以实现一种异常申请检测方法。
本领域技术人员可以理解,图7中示出的结构,仅仅是与本申请方案相关的部分结构的框图,并不构成对本申请方案所应用于其上的计算机设备的限定,具体的计算机设备可以包括比图中所示更多或更少的部件,或者组合某些部件,或者具有不同的部件布置。
一种计算机设备,包括存储器和一个或多个处理器,存储器中存储有计算机可读指令,计算机可读指令被处理器执行时实现本申请任意一个实施例中提供的异常申请检测方法的步骤。
一个或多个存储有计算机可读指令的非易失性存储介质,计算机可读指令被一个或多个处理器执行时,使得一个或多个处理器实现本申请任意一个 实施例中提供的异常申请检测方法方法的步骤。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程,是可以通过计算机可读指令来指令相关的硬件来完成,的计算机可读指令可存储于一非易失性计算机可读取存储介质中,该计算机可读指令在执行时,可包括如上述各方法的实施例的流程。其中,本申请所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用,均可包括非易失性和/或易失性存储器。非易失性存储器可包括只读存储器(ROM)、可编程ROM(PROM)、电可编程ROM(EPROM)、电可擦除可编程ROM(EEPROM)或闪存。易失性存储器可包括随机存取存储器(RAM)或者外部高速缓冲存储器。作为说明而非局限,RAM以多种形式可得,诸如静态RAM(SRAM)、动态RAM(DRAM)、同步DRAM(SDRAM)、双数据率SDRAM(DDRSDRAM)、增强型SDRAM(ESDRAM)、同步链路(Synchlink)DRAM(SLDRAM)、存储器总线(Rambus)直接RAM(RDRAM)、直接存储器总线动态RAM(DRDRAM)、以及存储器总线动态RAM(RDRAM)等。
以上实施例的各技术特征可以进行任意的组合,为使描述简洁,未对上述实施例中的各个技术特征所有可能的组合都进行描述,然而,只要这些技术特征的组合不存在矛盾,都应当认为是本说明书记载的范围。
以上实施例仅表达了本申请的几种实施方式,其描述较为具体和详细,但并不能因此而理解为对发明专利范围的限制。应当指出的是,对于本领域的普通技术人员来说,在不脱离本申请构思的前提下,还可以做出若干变形和改进,这些都属于本申请的保护范围。因此,本申请专利的保护范围应以所附权利要求为准。
Claims (20)
- 一种异常申请检测方法,包括:获取申请人的信贷申请数据;根据所述信贷申请数据获取申请人信息以及与所述申请人信息对应的申请特征集;将所述申请特征集输入预设孤立森林模型,通过所述预设孤立森林模型对所述申请特征集内特征进行异常检测,获取所述申请特征集对应的申请评分,所述预设孤立森林模型基于无标记的训练特征集构建,用于通过申请特征集内各个特征在所述孤立森林模型的孤立树中对应路径的长度进行异常特征检测,并基于所述路径的长度对申请特征集进行评分;及获取所述申请评分与预设基准值差值的绝对值,当所述差值的绝对值低于预设阈值时,判定所述信贷申请数据对应的申请为异常申请。
- 根据权利要求1所述的方法,其特征在于,所述根据所述信贷申请数据获取申请人信息以及与所述申请人信息对应的申请特征集,包括:根据所述信贷申请数据获取申请人信息;根据所述申请人信息,获取申请人的历史申请数据;及根据所述历史申请数据获得申请特征集。
- 根据权利要求1所述的方法,其特征在于,所述将申请特征集输入预设孤立森林模型,通过预设孤立森林模型对申请特征集内特征进行异常检测,获取申请特征集对应的申请评分包括:将所述申请特征集输入预设孤立森林模型;通过所述预设森林模型中的孤立树对申请特征集内的各个申请特征进行异常检测,获取各分别得到每个申请特征对应的异常度;及根据各申请特征的异常度,获取申请特征集对应的申请评分。
- 根据权利要求1所述的方法,其特征在于,所述将所述申请特征集输入预设孤立森林模型,获取所述申请特征集对应的申请评分之前,包括:通过预设无标记的训练特征集构建孤立树;及根据所述孤立树构建预设的孤立森林模型。
- 根据权利要求4所述的方法,其特征在于,所述通过预设无标记的训练特征集构建孤立树包括:从无标记的训练特征集中抽样出包含ψ个样本的样本特征集,作为孤立树的训练样本集;随机选取所述样本特征集内样本的单个特征;根据所述特征对所述样本特征集进行二叉划分,获得两个子特征集;判断所述子特征集是否能再次划分,当所述子特征集能再次划分时,将 子特征集作为新的样本特征集返回根据所述特征对所述样本特征集进行二叉划分,获得两个子特征集的步骤;及当所述子特征集内样本数等于1时,停止对所述子特征集的二叉划分。
- 根据权利要求5所述的方法,其特征在于,在根据所述特征对所述样本特征集进行二叉划分,获得两个子特征集之后,还包括:将子特征集作为新的样本特征集返回根据所述特征对所述样本特征集进行二叉划分,获得两个子特征集的步骤,记录返回次数,当所述返回次数大于log2(ψ)-2时,停止对所述子特征集的二叉划分。
- 根据权利要求6所述的方法,其特征在于,所述将所述申请特征集输入预设孤立森林模型,获取所述申请特征集对应的申请评分包括:获取所述申请特征集中申请特征;将对应的申请特征在所述预设孤立森林模型中的孤立树上运行,记录所述申请特征在所述运行过程中经过的路径长度;及根据所述申请特征集内所有申请特征对应的路径长度确定所述申请特征集的申请评分。
- 一种异常申请检测装置,包括:数据获取模块,用于获取申请人的信贷申请数据;特征提取模块,用于根据所述信贷申请数据获取申请人信息以及与所述申请人信息对应的申请特征集;评分计算模块,用于将所述申请特征集输入预设孤立森林模型,通过所述预设孤立森林模型对所述申请特征集内特征进行异常检测,获取所述申请特征集对应的申请评分,所述预设孤立森林模型基于无标记的训练特征集构建,用于通过申请特征集内各个特征在所述孤立森林模型的孤立树中对应路径的长度进行异常特征检测,并基于所述路径的长度对申请特征集进行评分;及异常申请判定模块,用于获取所述申请评分与预设基准值差值的绝对值,当所述差值的绝对值低于预设阈值时,判定所述信贷申请数据对应的申请为异常申请。
- 根据权利要求8所述的装置,其特征在于,所述特征提取模块具体用于:根据所述信贷申请数据获取申请人信息;根据所述申请人信息,获取申请人的历史申请数据;根据所述历史申请数据获得申请特征集。
- 一种计算机设备,包括存储器及一个或多个处理器,所述存储器中储存有计算机可读指令,所述计算机可读指令被所述一个或多个处理器执行 时,使得所述一个或多个处理器执行以下步骤:获取申请人的信贷申请数据;根据所述信贷申请数据获取申请人信息以及与所述申请人信息对应的申请特征集;将所述申请特征集输入预设孤立森林模型,通过所述预设孤立森林模型对所述申请特征集内特征进行异常检测,获取所述申请特征集对应的申请评分,所述预设孤立森林模型基于无标记的训练特征集构建,用于通过申请特征集内各个特征在所述孤立森林模型的孤立树中对应路径的长度进行异常特征检测,并基于所述路径的长度对申请特征集进行评分;及获取所述申请评分与预设基准值差值的绝对值,当所述差值的绝对值低于预设阈值时,判定所述信贷申请数据对应的申请为异常申请。
- 根据权利要求10所述的计算机设备,其特征在于,所述处理器执行所述计算机可读指令时还执行以下步骤:根据所述信贷申请数据获取申请人信息;根据所述申请人信息,获取申请人的历史申请数据;及根据所述历史申请数据获得申请特征集。
- 根据权利要求10所述的计算机设备,其特征在于,所述处理器执行所述计算机可读指令时还执行以下步骤:将所述申请特征集输入预设孤立森林模型;通过所述预设森林模型中的孤立树对申请特征集内的各个申请特征进行异常检测,获取各分别得到每个申请特征对应的异常度;及根据各申请特征的异常度,获取申请特征集对应的申请评分。
- 根据权利要求10所述的计算机设备,其特征在于,所述处理器执行所述计算机可读指令时还执行以下步骤:通过预设无标记的训练特征集构建孤立树;及根据所述孤立树构建预设的孤立森林模型。
- 根据权利要求13所述的计算机设备,其特征在于,所述处理器执行所述计算机可读指令时还执行以下步骤:从无标记的训练特征集中抽样出包含ψ个样本的样本特征集,作为孤立树的训练样本集;随机选取所述样本特征集内样本的单个特征;根据所述特征对所述样本特征集进行二叉划分,获得两个子特征集;判断所述子特征集是否能再次划分,当所述子特征集能再次划分时,将子特征集作为新的样本特征集返回根据所述特征对所述样本特征集进行二叉划分,获得两个子特征集的步骤;及当所述子特征集内样本数等于1时,停止对所述子特征集的二叉划分。
- 根据权利要求14所述的计算机设备,其特征在于,所述处理器执行所述计算机可读指令时还执行以下步骤:将子特征集作为新的样本特征集返回根据所述特征对所述样本特征集进行二叉划分,获得两个子特征集的步骤,记录返回次数,当所述返回次数大于log2(ψ)-2时,停止对所述子特征集的二叉划分。
- 一个或多个存储有计算机可读指令的非易失性计算机可读存储介质,所述计算机可读指令被一个或多个处理器执行时,使得所述一个或多个处理器执行以下步骤:获取申请人的信贷申请数据;根据所述信贷申请数据获取申请人信息以及与所述申请人信息对应的申请特征集;将所述申请特征集输入预设孤立森林模型,通过所述预设孤立森林模型对所述申请特征集内特征进行异常检测,获取所述申请特征集对应的申请评分,所述预设孤立森林模型基于无标记的训练特征集构建,用于通过申请特征集内各个特征在所述孤立森林模型的孤立树中对应路径的长度进行异常特征检测,并基于所述路径的长度对申请特征集进行评分;及获取所述申请评分与预设基准值差值的绝对值,当所述差值的绝对值低于预设阈值时,判定所述信贷申请数据对应的申请为异常申请。
- 根据权利要求16所述的存储介质,其特征在于,所述计算机可读指令被所述处理器执行时还执行以下步骤:根据所述信贷申请数据获取申请人信息;根据所述申请人信息,获取申请人的历史申请数据;及根据所述历史申请数据获得申请特征集。
- 根据权利要求16所述的存储介质,其特征在于,所述计算机可读指令被所述处理器执行时还执行以下步骤:将所述申请特征集输入预设孤立森林模型;通过所述预设森林模型中的孤立树对申请特征集内的各个申请特征进行异常检测,获取各分别得到每个申请特征对应的异常度;及根据各申请特征的异常度,获取申请特征集对应的申请评分。
- 根据权利要求16所述的存储介质,其特征在于,所述计算机可读指令被所述处理器执行时还执行以下步骤:通过预设无标记的训练特征集构建孤立树;及根据所述孤立树构建预设的孤立森林模型。
- 根据权利要求19所述的存储介质,其特征在于,所述计算机可读指 令被所述处理器执行时还执行以下步骤:从无标记的训练特征集中抽样出包含ψ个样本的样本特征集,作为孤立树的训练样本集;随机选取所述样本特征集内样本的单个特征;根据所述特征对所述样本特征集进行二叉划分,获得两个子特征集;判断所述子特征集是否能再次划分,当所述子特征集能再次划分时,将子特征集作为新的样本特征集返回根据所述特征对所述样本特征集进行二叉划分,获得两个子特征集的步骤;及当所述子特征集内样本数等于1时,停止对所述子特征集的二叉划分。
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910007045.4A CN109859029A (zh) | 2019-01-04 | 2019-01-04 | 异常申请检测方法、装置、计算机设备以及存储介质 |
| CN201910007045.4 | 2019-01-04 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2020140678A1 true WO2020140678A1 (zh) | 2020-07-09 |
Family
ID=66893857
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2019/123190 Ceased WO2020140678A1 (zh) | 2019-01-04 | 2019-12-05 | 异常申请检测方法、装置、计算机设备和存储介质 |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN109859029A (zh) |
| WO (1) | WO2020140678A1 (zh) |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111950853A (zh) * | 2020-07-14 | 2020-11-17 | 东南大学 | 一种基于信息物理双侧数据的电力运行状态白名单生成方法 |
| CN112733897A (zh) * | 2020-12-30 | 2021-04-30 | 胜斗士(上海)科技技术发展有限公司 | 确定多维样本数据的异常原因的方法和设备 |
| CN112971795A (zh) * | 2021-02-07 | 2021-06-18 | 中国人民解放军总医院 | 心电信号质量评估方法 |
| CN112971762A (zh) * | 2021-02-07 | 2021-06-18 | 中国人民解放军总医院 | 呼吸信号质量评估方法 |
| CN113076350A (zh) * | 2021-03-02 | 2021-07-06 | 无锡先导智能装备股份有限公司 | 焊接异常检测方法、装置、计算机设备和存储介质 |
| CN113283901A (zh) * | 2021-04-19 | 2021-08-20 | 河南大学 | 一种面向区块链平台的基于字节码的诈骗合约检测方法 |
| CN113537642A (zh) * | 2021-08-20 | 2021-10-22 | 日月光半导体制造股份有限公司 | 产品品质预测方法、装置、电子设备及存储介质 |
| CN114240059A (zh) * | 2021-11-22 | 2022-03-25 | 中国建设银行股份有限公司 | 资源在线申请处理方法、装置、计算机设备和存储介质 |
| CN114611616A (zh) * | 2022-03-16 | 2022-06-10 | 吕少岚 | 一种基于集成孤立森林的无人机智能故障检测方法及系统 |
| CN114979369A (zh) * | 2022-04-14 | 2022-08-30 | 马上消费金融股份有限公司 | 异常通话检测方法、装置、电子设备及存储介质 |
| CN115374436A (zh) * | 2021-10-19 | 2022-11-22 | 中国移动通信集团有限公司 | 异常检测方法、装置、设备及计算机可读存储介质 |
| CN115618068A (zh) * | 2022-10-10 | 2023-01-17 | 深圳依时货拉拉科技有限公司 | 道路数据检查方法及装置、计算机设备及可读存储介质 |
| CN116050455A (zh) * | 2022-11-07 | 2023-05-02 | 中国工商银行股份有限公司 | 异常用户识别方法、装置、设备及存储介质 |
| CN116561635A (zh) * | 2023-05-15 | 2023-08-08 | 中国工商银行股份有限公司 | 微服务架构下故障检测模型的训练方法、装置及设备 |
| CN116659826A (zh) * | 2022-08-23 | 2023-08-29 | 国家电投集团科学技术研究院有限公司 | 风电螺栓状态的检测方法及装置 |
| CN117312825A (zh) * | 2023-09-26 | 2023-12-29 | 航天信息股份有限公司 | 一种目标行为检测方法、装置、电子设备及存储介质 |
| CN117538491A (zh) * | 2024-01-09 | 2024-02-09 | 武汉怡特环保科技有限公司 | 一种站房空气质量智能监测方法及系统 |
| CN118113903A (zh) * | 2024-01-25 | 2024-05-31 | 腾讯音乐娱乐科技(深圳)有限公司 | 异常歌曲识别方法、装置、计算机设备和存储介质 |
| CN119179292A (zh) * | 2024-11-22 | 2024-12-24 | 江西省交通工程集团有限公司 | 一种混凝土拌合站信息自动采集方法及系统 |
| CN120524343A (zh) * | 2025-07-25 | 2025-08-22 | 泰安九洲金城机械有限公司 | 一种矿车发动机故障诊断方法及系统 |
Families Citing this family (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109859029A (zh) * | 2019-01-04 | 2019-06-07 | 深圳壹账通智能科技有限公司 | 异常申请检测方法、装置、计算机设备以及存储介质 |
| CN110378386B (zh) * | 2019-06-20 | 2024-09-06 | 平安科技(深圳)有限公司 | 基于有监督的无标记异常识别方法、装置及存储介质 |
| CN110288362A (zh) * | 2019-07-03 | 2019-09-27 | 北京工业大学 | 刷单的预测方法、装置及电子设备 |
| CN110322368A (zh) * | 2019-07-03 | 2019-10-11 | 厦门理工学院 | 一种谐波数据异常检测方法、终端设备及存储介质 |
| CN110398375B (zh) * | 2019-07-16 | 2021-10-19 | 广州亚美信息科技有限公司 | 车辆冷却系统工作状态的监测方法、装置、设备和介质 |
| CN110806546B (zh) * | 2019-10-28 | 2022-03-08 | 腾讯科技(深圳)有限公司 | 电池健康评估方法、装置、存储介质及电子设备 |
| CN113065610B (zh) * | 2019-12-12 | 2022-05-17 | 支付宝(杭州)信息技术有限公司 | 基于联邦学习的孤立森林模型构建和预测方法和装置 |
| CN112990246B (zh) * | 2019-12-17 | 2022-09-09 | 杭州海康威视数字技术股份有限公司 | 孤立树模型建立的方法和装置 |
| CN113032774B (zh) * | 2019-12-25 | 2024-06-07 | 中移动信息技术有限公司 | 异常检测模型的训练方法、装置、设备及计算机存储介质 |
| CN111340063B (zh) * | 2020-02-10 | 2023-08-29 | 国能信控互联技术有限公司 | 一种磨煤机数据异常检测方法 |
| CN111275547B (zh) * | 2020-03-19 | 2023-07-18 | 重庆富民银行股份有限公司 | 基于孤立森林的风控系统及方法 |
| CN111612040B (zh) * | 2020-04-24 | 2024-04-30 | 平安直通咨询有限公司上海分公司 | 基于孤立森林算法的财务数据异常检测方法及相关装置 |
| CN111833172A (zh) * | 2020-05-25 | 2020-10-27 | 百维金科(上海)信息科技有限公司 | 一种基于孤立森林的消费信贷欺诈行为检测方法及其系统 |
| CN113159923A (zh) * | 2021-04-29 | 2021-07-23 | 中国工商银行股份有限公司 | 风险筛查方法及装置 |
| CN115689407A (zh) * | 2021-07-29 | 2023-02-03 | 顺丰科技有限公司 | 账户异常检测方法、装置以及终端设备 |
| CN113961434A (zh) * | 2021-09-29 | 2022-01-21 | 西安交通大学 | 一种分布式区块链系统用户异常行为监测方法及系统 |
| CN114580580B (zh) * | 2022-05-07 | 2022-08-16 | 深圳索信达数据技术有限公司 | 一种智能运维异常检测方法及装置 |
| CN115065597B (zh) * | 2022-06-14 | 2024-06-07 | 中国电信股份有限公司 | 一种容器资源配置方法、装置、介质和设备 |
| CN116166651A (zh) * | 2023-02-15 | 2023-05-26 | 平安科技(深圳)有限公司 | 客户端数据质量监控方法、装置、设备及介质 |
| CN117786543B (zh) * | 2024-02-28 | 2024-05-10 | 沂水友邦养殖服务有限公司 | 一种数字化肉鸡养殖信息存储管理方法及系统 |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2016200971A (ja) * | 2015-04-09 | 2016-12-01 | キヤノン株式会社 | 学習装置、識別装置、学習方法、識別方法及びプログラム |
| CN107357790A (zh) * | 2016-05-09 | 2017-11-17 | 阿里巴巴集团控股有限公司 | 一种异常消息检测方法、装置及系统 |
| CN107391569A (zh) * | 2017-06-16 | 2017-11-24 | 阿里巴巴集团控股有限公司 | 数据类型的识别、模型训练、风险识别方法、装置及设备 |
| CN108549973A (zh) * | 2018-03-22 | 2018-09-18 | 中国平安人寿保险股份有限公司 | 识别模型构建及评估的方法、装置、存储介质及终端 |
| CN109859029A (zh) * | 2019-01-04 | 2019-06-07 | 深圳壹账通智能科技有限公司 | 异常申请检测方法、装置、计算机设备以及存储介质 |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060218079A1 (en) * | 2005-02-08 | 2006-09-28 | Goldblatt Joel N | Web-based consumer loan database with automated controls for preventing predatory lending practices |
| US8484132B1 (en) * | 2012-06-08 | 2013-07-09 | Lexisnexis Risk Solutions Fl Inc. | Systems and methods for segmented risk scoring of identity fraud |
| CN106960358A (zh) * | 2017-01-13 | 2017-07-18 | 重庆小富农康农业科技服务有限公司 | 一种基于农村电子商务大数据深度学习的金融欺诈行为量化检测系统 |
| CN107886425A (zh) * | 2017-10-25 | 2018-04-06 | 上海壹账通金融科技有限公司 | 信贷评估方法、装置、设备及计算机可读存储介质 |
| CN108038700A (zh) * | 2017-12-22 | 2018-05-15 | 上海前隆信息科技有限公司 | 一种反欺诈数据分析方法与系统 |
| CN108364224A (zh) * | 2018-01-12 | 2018-08-03 | 深圳壹账通智能科技有限公司 | 信贷风险联控方法、装置、设备及可读存储介质 |
| CN108985632A (zh) * | 2018-07-16 | 2018-12-11 | 国网上海市电力公司 | 一种基于孤立森林算法的用电数据异常检测模型 |
-
2019
- 2019-01-04 CN CN201910007045.4A patent/CN109859029A/zh active Pending
- 2019-12-05 WO PCT/CN2019/123190 patent/WO2020140678A1/zh not_active Ceased
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2016200971A (ja) * | 2015-04-09 | 2016-12-01 | キヤノン株式会社 | 学習装置、識別装置、学習方法、識別方法及びプログラム |
| CN107357790A (zh) * | 2016-05-09 | 2017-11-17 | 阿里巴巴集团控股有限公司 | 一种异常消息检测方法、装置及系统 |
| CN107391569A (zh) * | 2017-06-16 | 2017-11-24 | 阿里巴巴集团控股有限公司 | 数据类型的识别、模型训练、风险识别方法、装置及设备 |
| CN108549973A (zh) * | 2018-03-22 | 2018-09-18 | 中国平安人寿保险股份有限公司 | 识别模型构建及评估的方法、装置、存储介质及终端 |
| CN109859029A (zh) * | 2019-01-04 | 2019-06-07 | 深圳壹账通智能科技有限公司 | 异常申请检测方法、装置、计算机设备以及存储介质 |
Cited By (27)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111950853B (zh) * | 2020-07-14 | 2024-05-31 | 东南大学 | 一种基于信息物理双侧数据的电力运行状态白名单生成方法 |
| CN111950853A (zh) * | 2020-07-14 | 2020-11-17 | 东南大学 | 一种基于信息物理双侧数据的电力运行状态白名单生成方法 |
| CN112733897A (zh) * | 2020-12-30 | 2021-04-30 | 胜斗士(上海)科技技术发展有限公司 | 确定多维样本数据的异常原因的方法和设备 |
| CN112971795A (zh) * | 2021-02-07 | 2021-06-18 | 中国人民解放军总医院 | 心电信号质量评估方法 |
| CN112971762A (zh) * | 2021-02-07 | 2021-06-18 | 中国人民解放军总医院 | 呼吸信号质量评估方法 |
| CN112971762B (zh) * | 2021-02-07 | 2023-04-18 | 中国人民解放军总医院 | 呼吸信号质量评估方法 |
| CN112971795B (zh) * | 2021-02-07 | 2023-04-18 | 中国人民解放军总医院 | 心电信号质量评估方法 |
| CN113076350B (zh) * | 2021-03-02 | 2024-05-07 | 无锡先导智能装备股份有限公司 | 焊接异常检测方法、装置、计算机设备和存储介质 |
| CN113076350A (zh) * | 2021-03-02 | 2021-07-06 | 无锡先导智能装备股份有限公司 | 焊接异常检测方法、装置、计算机设备和存储介质 |
| CN113283901A (zh) * | 2021-04-19 | 2021-08-20 | 河南大学 | 一种面向区块链平台的基于字节码的诈骗合约检测方法 |
| CN113537642A (zh) * | 2021-08-20 | 2021-10-22 | 日月光半导体制造股份有限公司 | 产品品质预测方法、装置、电子设备及存储介质 |
| CN115374436A (zh) * | 2021-10-19 | 2022-11-22 | 中国移动通信集团有限公司 | 异常检测方法、装置、设备及计算机可读存储介质 |
| CN114240059A (zh) * | 2021-11-22 | 2022-03-25 | 中国建设银行股份有限公司 | 资源在线申请处理方法、装置、计算机设备和存储介质 |
| CN114611616B (zh) * | 2022-03-16 | 2023-02-07 | 吕少岚 | 一种基于集成孤立森林的无人机智能故障检测方法及系统 |
| CN114611616A (zh) * | 2022-03-16 | 2022-06-10 | 吕少岚 | 一种基于集成孤立森林的无人机智能故障检测方法及系统 |
| CN114979369A (zh) * | 2022-04-14 | 2022-08-30 | 马上消费金融股份有限公司 | 异常通话检测方法、装置、电子设备及存储介质 |
| CN116659826A (zh) * | 2022-08-23 | 2023-08-29 | 国家电投集团科学技术研究院有限公司 | 风电螺栓状态的检测方法及装置 |
| CN116659826B (zh) * | 2022-08-23 | 2024-02-06 | 国家电投集团科学技术研究院有限公司 | 风电螺栓状态的检测方法及装置 |
| CN115618068A (zh) * | 2022-10-10 | 2023-01-17 | 深圳依时货拉拉科技有限公司 | 道路数据检查方法及装置、计算机设备及可读存储介质 |
| CN116050455A (zh) * | 2022-11-07 | 2023-05-02 | 中国工商银行股份有限公司 | 异常用户识别方法、装置、设备及存储介质 |
| CN116561635A (zh) * | 2023-05-15 | 2023-08-08 | 中国工商银行股份有限公司 | 微服务架构下故障检测模型的训练方法、装置及设备 |
| CN117312825A (zh) * | 2023-09-26 | 2023-12-29 | 航天信息股份有限公司 | 一种目标行为检测方法、装置、电子设备及存储介质 |
| CN117538491A (zh) * | 2024-01-09 | 2024-02-09 | 武汉怡特环保科技有限公司 | 一种站房空气质量智能监测方法及系统 |
| CN117538491B (zh) * | 2024-01-09 | 2024-04-05 | 武汉怡特环保科技有限公司 | 一种站房空气质量智能监测方法及系统 |
| CN118113903A (zh) * | 2024-01-25 | 2024-05-31 | 腾讯音乐娱乐科技(深圳)有限公司 | 异常歌曲识别方法、装置、计算机设备和存储介质 |
| CN119179292A (zh) * | 2024-11-22 | 2024-12-24 | 江西省交通工程集团有限公司 | 一种混凝土拌合站信息自动采集方法及系统 |
| CN120524343A (zh) * | 2025-07-25 | 2025-08-22 | 泰安九洲金城机械有限公司 | 一种矿车发动机故障诊断方法及系统 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109859029A (zh) | 2019-06-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2020140678A1 (zh) | 异常申请检测方法、装置、计算机设备和存储介质 | |
| US20210092160A1 (en) | Data set creation with crowd-based reinforcement | |
| CN108509485B (zh) | 数据的预处理方法、装置、计算机设备和存储介质 | |
| US9965531B2 (en) | Data storage extract, transform and load operations for entity and time-based record generation | |
| US20180253657A1 (en) | Real-time credit risk management system | |
| Banerjee | Population growth and endogenous technological change: Australian economic growth in the long run | |
| Lima et al. | Domain knowledge integration in data mining using decision tables: case studies in churn prediction | |
| CN114579584B (zh) | 数据表处理方法、装置、计算机设备和存储介质 | |
| WO2020211299A1 (zh) | 数据清理方法 | |
| WO2019061994A1 (zh) | 电子装置、保险产品推荐方法、系统及计算机可读存储介质 | |
| WO2019218699A1 (zh) | 欺诈交易判断方法、装置、计算机设备和存储介质 | |
| CN111784392A (zh) | 基于孤立森林的异常用户群组检测方法、装置、设备 | |
| CN114495137B (zh) | 票据异常检测模型生成方法与票据异常检测方法 | |
| WO2017133615A1 (zh) | 一种业务参数获取方法及装置 | |
| CN112035611B (zh) | 目标用户推荐方法、装置、计算机设备和存储介质 | |
| CN114037478A (zh) | 广告异常流量检测方法、系统、电子设备及可读存储介质 | |
| WO2018184518A1 (zh) | 微博数据处理方法、装置、计算机设备及存储介质 | |
| CN111582932A (zh) | 场景间信息推送方法、装置、计算机设备及存储介质 | |
| CN114529399A (zh) | 用户数据处理方法、装置、计算机设备和存储介质 | |
| CN119886084A (zh) | 一种数据表合并拆分方法、装置、设备、介质及程序产品 | |
| CN110727711B (zh) | 基金数据库中异常数据检测方法、装置和计算机设备 | |
| CN116155597A (zh) | 访问请求的处理方法、装置及计算机设备 | |
| CN111127057A (zh) | 一种多维用户画像恢复方法 | |
| US20240223615A1 (en) | System and method for data set creation with crowd-based reinforcement | |
| CN118740662A (zh) | 异常请求检测方法、装置、计算机设备和存储介质 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 19907781 Country of ref document: EP Kind code of ref document: A1 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 32PN | Ep: public notification in the ep bulletin as address of the adressee cannot be established |
Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 20/10/2021) |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 19907781 Country of ref document: EP Kind code of ref document: A1 |