Early warning method and device for preventing customer loss, equipment and medium

Abstract

The invention provides an early warning method for preventing customer loss. The early warning method can be applied to the field of big data or artificial intelligence. The method comprises the following steps: under the condition that a customer agrees or authorizes, acquiring m pieces of business data stored in a bank business system by the customer; data analysis is conducted on the m pieces of service data, n1 behavior path characteristics and n2 event characteristics are obtained, the behavior path characteristics are used for representing behavior paths of the vulnerable customers, and the event characteristics are used for representing events of the vulnerable customers; generating n3 loss features according to the n1 behavior path features and the n2 event features; based on the n3 churn features, predicting the churn probability of the customer by using a pre-trained churn prediction model; and based on the loss probability, carrying out risk grade division on the customer, and carrying out corresponding early warning on the customer according to the risk grade. The invention further provides an early warning device for preventing customer loss, equipment, a storage medium and a program product.

Description

Technical Field

[0001] The present disclosure relates to the field of big data or artificial intelligence, and more specifically to an early warning method, device, equipment, medium and program product for preventing customer churn. Background Art

[0002] In the modern financial field, commercial banks, as important providers of financial services, have a huge personal customer base, which can easily reach hundreds of millions, and the amount of business data stored in the banking system for each customer is also large.

[0003] In the traditional operation method, it is difficult to timely and accurately identify whether a customer will churn by relying on the personal experience of the account manager to maintain the customer. With the development of technology, some big data analysis or artificial intelligence technologies have begun to be used in related technologies to identify whether a customer will churn. However, these big data analysis or artificial intelligence technologies usually use the original business data of the customer, which occupies a large storage space and is not conducive to saving storage space; moreover, during the processing, a large amount of original business data needs to be circulated and processed, which will lead to unreasonable occupation of the transmission bandwidth of the banking business system and reduce the processing speed; in addition, using a large amount of original business data to identify whether a customer will churn has the problem of low recognition accuracy.

[0004] Therefore, the current operating methods make it difficult to timely and accurately identify whether a customer is about to churn. Waiting until the customer has already churned before taking retention measures is often too late, causing the bank to lose valuable customer resources. Summary of the invention

[0005] In view of the above problems, embodiments of the present disclosure provide an early warning method, apparatus, device, medium and program product for preventing customer churn.

[0006] According to a first aspect of the present disclosure, there is provided an early warning method for preventing customer churn, the method comprising: obtaining, with the consent or authorization of the customer, m business data of the customer stored in a banking business system; performing data analysis on the m business data to obtain n1 behavior path features and n2 event features, wherein the behavior path features are used to characterize the behavior paths of customers prone to churn, and the event features are used to characterize events occurring to customers prone to churn, and the storage space occupied by the n1 behavior path features and the n2 event features is smaller than the storage space occupied by the m business data; generating n3 churn features based on the n1 behavior path features and the n2 event features, wherein the storage space occupied by the n3 churn features is smaller than the storage space occupied by the n1 behavior path features and the n2 event features; predicting the churn probability of the customer using a pre-trained churn prediction model based on the n3 churn features; and classifying the customer into risk levels based on the churn probability, and issuing corresponding early warnings to the customer according to the risk levels.

[0007] According to an embodiment of the present disclosure, the method of predicting the churn probability of the customer based on the n3 churn features using a pre-trained churn prediction model includes: taking the n3 churn features as input, respectively inputting them into k pre-built primary learners to obtain k primary output variables, wherein k is a positive integer greater than or equal to 2; and taking the k primary output variables as input, inputting them into a pre-built secondary learner to obtain the churn probability of the customer, wherein the secondary learner serves as a fusion model to fuse the output variables of the k primary learners.

[0008] According to an embodiment of the present disclosure, generating n3 churn features based on the n1 behavior path features and the n2 event features includes: using a filtering method to filter out n4 primary churn features from the n1 behavior path features and the n2 event features, wherein n4 is less than the sum of n1 and n2 and n4 is greater than n3; and using an embedded method to filter out n3 churn features from the n4 primary churn features.

[0009] According to an embodiment of the present disclosure, the method further includes: based on the n3 churn features, outputting a feature importance list using a pre-trained churn prediction model, wherein the feature importance list includes impact values ​​of the n3 churn features, and the impact values ​​of the n3 churn features are used to characterize the importance of the impact of the n3 churn features on the churn probability of the customer.

[0010] According to an embodiment of the present disclosure, the m business data include the asset data of the customer, and the data analysis of the m business data to obtain n1 behavior path characteristics and n2 event characteristics includes: determining the financial behavior of the customer in the current time period based on the asset data of the customer, and dividing the customer into at least one of channel dependence, savings, investment and net expenditure types based on the financial behavior to obtain the financial status of the customer; and establishing an observation period of a preset time period for the customer based on the financial status of the customer, and analyzing changes in the financial behavior of the customer during the observation period to determine the behavior path characteristics of the customer.

[0011] According to an embodiment of the present disclosure, the m business data are subjected to data analysis to obtain n1 behavior path features and n2 event features, including: parsing the customer's transaction identifier based on the customer's asset data, and identifying the customer's transaction event based on the transaction identifier, wherein the transaction identifier is used to determine the transaction information of the customer executing the transaction event; classifying the customer's transaction events to form important events; and forming the event features based on the important events.

[0012] According to an embodiment of the present disclosure, training the churn prediction model includes: obtaining the business data of the customer that meets the preset analysis conditions within a historical time period, and performing data analysis to obtain the behavior path characteristics and the event characteristics; using a filtering method and an embedded method to perform feature screening on the behavior path characteristics and the event characteristics to obtain the churn characteristics; inputting the churn characteristics into a plurality of pre-built primary learners, and obtaining the primary output variables output by each of the primary learners; inputting each of the primary output variables into a pre-built secondary learner to obtain the churn probability output by the secondary learner; and using a statistical test method to test the churn probability output by the secondary learner, and adjusting the parameters of the churn prediction model according to the statistical value calculated by the statistical test method until the statistical distribution of the churn probability is consistent with the statistical distribution of the actual churn situation.

[0013] A second aspect of the present disclosure provides an early warning device for preventing customer churn, comprising: an acquisition module, for acquiring m business data of a customer stored in a banking business system with the customer's consent or authorization; an analysis module, for performing data analysis on the m business data to acquire n1 behavior path features and n2 event features, wherein the behavior path features are used to characterize the behavior paths of customers prone to churn, and the event features are used to characterize events occurring to customers prone to churn, and the storage space occupied by the n1 behavior path features and the n2 event features is smaller than the storage space occupied by the m business data; a generation module, for generating n3 churn features based on the n1 behavior path features and the n2 event features, wherein the storage space occupied by the n3 churn features is smaller than the storage space occupied by the n1 behavior path features and the n2 event features; a prediction module, for predicting the churn probability of the customer using a pre-trained churn prediction model based on the n3 churn features; and an early warning module, for classifying the churn customers into risk levels based on the churn probability, and issuing corresponding early warnings to the customers according to the risk levels.

[0014] A third aspect of the present disclosure provides an electronic device, comprising: one or more processors; a storage device for storing one or more programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the above-mentioned early warning method for preventing customer churn.

[0015] The fourth aspect of the present disclosure further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-mentioned early warning method for preventing customer churn.

[0016] The fifth aspect of the present disclosure further provides a computer program product, including a computer program, which implements the steps of the above-mentioned early warning method for preventing customer churn when executed by a processor.

[0017] In the embodiments of the present disclosure, big data is used to analyze factors affecting customer churn, determine customer behavior path characteristics and event characteristics, and further obtain churn characteristics and incorporate them into a churn prediction model. The churn prediction model is used to predict the customer churn probability, and customers are classified into risk levels based on the churn probability. Customers with signs of churn are warned in advance, thereby reasonably occupying the transmission bandwidth of the banking business system, saving storage space, improving processing speed and recognition accuracy, and helping business personnel to optimize customer service in a timely manner, carry out targeted retention marketing, and improve individual customer quality and capital retention rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The above contents and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0019] Figure 1 The application scenario diagram schematically shows the early warning method, device, equipment, medium and program product for preventing customer churn according to the embodiment of the present disclosure;

[0020] Figure 2 A flowchart schematically illustrates an early warning method for preventing customer churn according to an embodiment of the present disclosure;

[0021] Figure 3 A schematic diagram schematically shows the classification of customer financial status according to an embodiment of the present disclosure;

[0022] Figure 4 A flowchart of parsing a transaction identifier according to an embodiment of the present disclosure is schematically shown;

[0023] Figure 5 A flowchart of application monitoring rules according to an embodiment of the present disclosure is schematically shown;

[0024] Figure 6 A structural block diagram of an early warning device for preventing customer loss according to an embodiment of the present disclosure is schematically shown; and

[0025] Figure 7 A block diagram schematically shows an electronic device for an early warning method for preventing customer churn according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0026] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0027] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0028] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0029] When using expressions such as "at least one of A, B, and C, etc.", they should generally be interpreted according to the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0030] In the technical solution of the present invention, the user information (including but not limited to user personal information, user image information, user device information, such as location information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with the relevant laws, regulations and standards of the relevant countries and regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0031] Figure 1 The application scenario diagram of the early warning method, device, equipment, medium and program product for preventing customer churn according to the embodiments of the present disclosure is schematically shown.

[0032] like Figure 1 As shown, the application scenario 100 according to this embodiment may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for a communication link between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0033] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only examples).

[0034] The terminal devices 101 , 102 , and 103 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, and desktop computers.

[0035] The server 105 may be a server that provides various services, such as a background management server (only as an example) that provides support for websites browsed by users using the terminal devices 101, 102, and 103. The background management server may analyze and process the received data such as user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal device.

[0036] It should be noted that the early warning method for preventing customer loss provided in the embodiment of the present disclosure can generally be executed by the server 105. Accordingly, the early warning device for preventing customer loss provided in the embodiment of the present disclosure can generally be set in the server 105. The early warning method for preventing customer loss provided in the embodiment of the present disclosure can also be executed by a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105. Accordingly, the early warning device for preventing customer loss provided in the embodiment of the present disclosure can also be set in a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105.

[0037] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to the implementation requirements.

[0038] Embodiments of the present disclosure provide an early warning method, apparatus, device, medium, and program product for preventing customer churn.

[0039] The following will be based on Figure 1 The scene described by Figure 2-Figure 5 The early warning method for preventing customer loss according to an embodiment of the present disclosure is described in detail.

[0040] In the disclosed embodiment, the early warning method for preventing customer churn is divided into two parts: data analysis and model application. The data analysis includes behavioral path features and event features related to customer churn, and finally feature screening to obtain churn features. The model application includes introducing the churn features mined in the previous step of "data analysis", predicting the churn probability through a machine learning algorithm, i.e., a churn prediction model, and mining important features of customer churn.

[0041] In an embodiment of the present disclosure, training the churn prediction model includes: obtaining the business data of the customer that meets the preset analysis conditions within a historical time period, and performing data analysis to obtain the behavior path characteristics and the event characteristics; using a filtering method and an embedded method to perform feature screening on the behavior path characteristics and the event characteristics to obtain the churn characteristics; inputting the churn characteristics into a plurality of pre-built primary learners, and obtaining the primary output variables output by each of the primary learners; inputting each of the primary output variables into a pre-built secondary learner to obtain the churn probability output by the secondary learner; and using a statistical test method to test the churn probability output by the secondary learner, and adjusting the parameters of the churn prediction model according to the statistical value calculated by the statistical test method until the statistical distribution of the churn probability conforms to the statistical distribution of the actual churn situation.

[0042] Specifically, the analysis object is first determined, that is, the customer who meets the preset analysis conditions in the historical time period. In some exemplary embodiments, the customer who meets the preset analysis conditions is the customer whose average monthly daily assets in months A, A+1, and A+2 are greater than 50,000. Then the basis point is calculated, basis point (average monthly daily assets) = (average daily assets in month A+1+average daily assets in month A+2) / 3. Then determine whether it is a lost customer. The lost customer (Y=1) meets the conditions that the monthly daily average asset balance for 4 consecutive months from month A+3 to month A+6 is < basis point*50%; the non-lost customer (Y=0) meets the conditions that the monthly daily average asset balance for at least one month from month A+3 to month A+6 is ≧ basis point*50%. In this process, the monthly base table of the training model is from month A to month A+6, and the feature data table of the model training is from month A-3 to month A+3.

[0043] Furthermore, the acquired data is preprocessed, including duplicate value processing, that is, deduplication by customer ID, and missing value processing, that is, checking the number of missing values ​​for each indicator, deleting indicators with a missing ratio exceeding a certain percentage (such as 60%), and at the same time, checking the number of missing indicators for each sample, deleting samples with a high proportion of missing indicators, and controlling the deleted samples to a certain proportion (such as 2%) of the total number of samples. In addition, outlier processing is also required, such as selecting continuous type fields, sorting them from small to large values, deleting samples exceeding μ±3σ (or a certain percentage before and after), and controlling the total number of deleted samples to a certain proportion (such as 5%) of the total number of samples. Alternatively, the K-means clustering algorithm is used for binning continuous type fields, and outliers will be automatically clustered into one category. During the training process, the model will examine the impact of outliers on whether customers churn.

[0044] After preprocessing, data analysis is performed to obtain behavior path characteristics and event characteristics, namely, constructed variables. The construction rules of event class variables are shown in Table 1.

[0045] Table 1:

[0046]

[0047] For example, the construction rules of the monthly cumulative table variables in the general variables are shown in Table 2.

[0048] Table 2: Monthly accumulation table variable construction rules

[0049]

[0050] By classifying and summarizing business data, such as customer account balances, transaction frequency, loan status, and financial product purchase records, we can better understand the customer's asset status, consumption habits, and risk preferences, and provide basic data for subsequent churn prediction. Then compare the data of different time periods to find the changing trends of customers in terms of account balances and transaction frequency. On the basis of identifying the direction of trend changes, further analyze whether these changes indicate the risk of customer churn. For example, if a customer's account balance continues to decline and the transaction frequency also decreases significantly, this may mean that the customer's satisfaction with the bank is declining and there is a possibility of churn. Finally, cross-analyze different types of indicators to obtain more comprehensive and in-depth insights into customer churn. For example, cross-analyze the customer's account balance and loan status to find out whether there is a situation where customer churn is caused by excessive loan pressure. Through this series of analysis and mining processes, derived variables can be obtained, that is, features related to behavioral path characteristics and important event characteristics. This variable not only takes into account the customer's various behavioral characteristics and transaction patterns, but also combines the time factor, which can more accurately reflect the customer's churn risk.

[0051] After obtaining the behavior path features and event features, the churn features are obtained through a two-stage feature selection strategy (filtering method and embedded method), and then the churn features are successively input into the primary learner and the secondary learner to obtain the predicted churn probability, for example, output a binary variable, 0 means that the customer does not meet the churn criteria, and 1 means that the customer meets the churn criteria. Finally, the difference between the predicted output and the actual churn probability is calculated (such as evaluation indicators such as accuracy, recall rate and F1 score), and the statistical test method is used to test the model's ability to distinguish customer churn risks and quantify the performance of the model. The statistical value is an evaluation indicator commonly used in binary classification problems, which measures the model's ability to distinguish between positive and negative samples. The higher the statistical value, the stronger the model's ability to distinguish.

[0052] It should be noted that when the model performance is poor, the model needs to be adjusted. There are many ways to adjust the model, including but not limited to: adjusting model parameters, such as learning rate, number of iterations, and regularization coefficient, to optimize the performance of the model, or adding or deleting some features based on feature importance analysis or business logic to improve the model's predictive ability.

[0053] It is understandable that by using multiple primary learners to train the model, making full use of the advantages of each model and reducing the deviation of a single model, it is possible to capture multiple features and patterns of the data. At the same time, the secondary learners are used to further optimize and combine the prediction results of each primary learner to improve the stability and generalization ability of the overall model. In addition, through the statistical test method, the prediction effect of the model on customer churn risk can be objectively evaluated to ensure the effectiveness and reliability of the model in predicting customer churn risk in practical applications.

[0054] Figure 2 The flowchart of the early warning method for preventing customer churn according to an embodiment of the present disclosure is schematically shown.

[0055] like Figure 2 As shown, the early warning method for preventing customer loss in this embodiment includes operations S210 to S250.

[0056] In operation S210, with the consent or authorization of the customer, m business data of the customer stored in the banking business system are obtained.

[0057] In an embodiment of the present disclosure, before acquiring m business data stored in a banking business system by a customer, the user's consent or authorization is obtained. For example, before operation S210, a request to acquire business data is issued to the user. When the user agrees or authorizes that the business data can be acquired, operation S210 is performed.

[0058] In operation S220, data analysis is performed on the m business data to obtain n1 behavior path features and n2 event features, wherein the behavior path features are used to characterize the behavior paths of customers who are prone to churn, and the event features are used to characterize events that occur to customers who are prone to churn, and the storage space occupied by the n1 behavior path features and the n2 event features is smaller than the storage space occupied by the m business data.

[0059] In operation S230, n3 churn features are generated according to the n1 behavior path features and the n2 event features, wherein the storage space occupied by the n3 churn features is smaller than the storage space occupied by the n1 behavior path features and the n2 event features. A churn prediction model is pre-trained.

[0060] In operation S240, based on the n3 churn features, a pre-trained churn prediction model is used to predict the churn probability of the customer.

[0061] In the embodiment of the present disclosure, a corresponding operation entry is provided for the user to choose to agree or reject the automated decision result. That is, before the churn probability processing / decision of the churned customer is predicted using the pre-trained churn prediction model, an instruction of the user to agree or reject the processing / decision is obtained through the corresponding operation entry. If the user agrees to perform the processing / decision, the churn probability processing / decision of the churned customer is predicted using the pre-trained churn prediction model, that is, step S240 is executed. If the user refuses to perform the processing / decision, the expert decision process is entered.

[0062] In operation S250, the lost customers are classified into risk levels based on the loss probability, and corresponding warnings are issued to the customers according to the risk levels.

[0063] Exemplarily, the word "churn" can have different meanings in different contexts, and different definitions will also determine the direction of the analysis. To ensure rigor, in some exemplary embodiments, churn can be defined as: taking the average monthly daily asset balance from the 1st to the 3rd month as the base point, starting from the 4th month, the customer's monthly average daily asset balance is compared with the base point value every month. If the customer's monthly average daily asset balance in the next 4 months does not reach more than 50% of the base point value, the customer is defined as a churned customer.

[0064] Specifically, the business data of connected customers is obtained from the bank's business system. This data may include the customer's deposit balance, investment product holdings, and transaction frequency. By analyzing these business data, the bank can preliminarily determine which customers are at risk of churn, and then further analyze the business data of these customers to determine the behavioral path characteristics and event characteristics that lead to their churn, such as sudden fund transfers, frequent transaction anomalies, and redemption of investment products, so as to gain a deeper understanding of the reasons for customer churn and provide an important basis for subsequent feature screening and churn prediction. Then, through feature screening technology, churn features that can reflect customer churn risk are extracted from behavioral path features and event features. Then, based on data analysis, the churn prediction model constructed by the machine learning algorithm specifically predicts the probability of customer churn, and sorts the prediction results from high to low to determine the priority of recovery.

[0065] It is understandable that by conducting group classification analysis on customer behavior, mining customer behavior path characteristics and event characteristics, and summarizing and refining churn characteristics and incorporating them into the churn prediction model, it is helpful to reasonably occupy the transmission bandwidth of the banking business system, save storage space, improve processing speed and recognition accuracy, and at the same time, assist commercial banks to further understand the deep reasons for customer churn and promote the optimization of customer service. In addition, the early warning method for preventing customer churn provided by the embodiment of the present disclosure has achieved good results in practical applications, and can be flexibly promoted and replicated to other commercial banks, helping to promote high-quality development of the banking industry.

[0066] Furthermore, the m business data include the asset data of the customer, and the data analysis of the m business data to obtain n1 behavior path characteristics and n2 event characteristics includes: determining the financial behavior of the customer in the current time period based on the asset data of the customer, and dividing the customer into at least one of channel dependence, savings, investment and net expenditure types according to the financial behavior to obtain the financial status of the customer; and establishing an observation period of a preset time period for the customer based on the financial status of the customer, and analyzing changes in the customer's financial behavior during the observation period to determine the customer's behavior path characteristics.

[0067] Figure 3 A schematic diagram of the classification of customer financial status according to an embodiment of the present disclosure is schematically shown.

[0068] like Figure 3 As shown, in some exemplary embodiments, the financial behavior of the customer in the current time period (such as the current month) can be determined by analyzing the asset data in the business data. For example, according to whether the individual customer has channel dependence, whether he holds regular savings products, whether he holds investment products, and whether he has net spending behavior in the current time period, the customer's financial status can be defined as follows, describing the customer as channel dependence, savings, investment, and net spending. Then, according to the customer's financial behavior in the current time period, the customer is divided into four categories and 16 financial statuses, as shown in Table 3:

[0069] Table 3

[0070]

[0071] For example, based on the financial status of customers each month, an observation period of a preset time period is established for lost customers, such as 6 months. The time of customer loss is pushed forward 6 months, and the key paths of changes in the customer's financial behavior during the 6-month observation period are reviewed.

[0072] By analyzing the changes in the financial behavior of lost customers, we found three typical behavioral path characteristics: stable low demand type, unstable low demand type, and loss type. Among them, stable low demand type means that the customer behavior label (such as investment redemption, fund loss, and channel separation) during the observation period is less than or equal to 2 and the customer's financial behavior has not changed during the entire observation period; unstable low demand type means that the customer behavior label is less than or equal to 2 during the observation period and the customer's financial behavior fluctuates during the observation period; loss type means that during the observation period, the customer's behavior at the end of the period is worse than the behavior at the beginning of the period.

[0073] In some exemplary embodiments, it can be observed that the customer churn process is a behavioral path of "investment redemption-fund loss-channel exit", and has the following characteristics: the churn warning effect of changes in time deposit behavior is not obvious; investment behavior sees a redemption peak 4 months before churn; net fund outflow behavior mainly starts 2-3 months before churn; channel exit behavior is mainly manifested 1-3 months before churn.

[0074] It is understandable that by classifying customers into different financial statuses, banks can more accurately understand customers' financial needs and risk preferences, thereby more accurately identifying customers' behavioral path characteristics, and then identifying the key nodes and reasons for customer churn, providing support for subsequent early warning and retention work.

[0075] Furthermore, the data analysis is performed on the m business data to obtain n1 behavior path features and n2 event features, including: parsing the customer's transaction identifier based on the customer's asset data, and identifying the customer's transaction event based on the transaction identifier, wherein the transaction identifier is used to determine the transaction information of the customer executing the transaction event; classifying the customer's transaction events to form important events; and forming the event features based on the important events.

[0076] Specifically, based on the analysis of customer behavior paths, we can identify customer transaction events by parsing transaction identifiers (transaction codes), and classify events with higher average customer frequency and events with higher average average customer frequency separately to form important events, and based on important events, form event characteristics.

[0077] Figure 4 The flowchart of parsing a transaction identifier according to an embodiment of the present disclosure is schematically shown.

[0078] like Figure 4As shown, after parsing the transaction identifier, i.e., the transaction code, important events can be divided into Class A events, Class B events, and Class C events. Each type of event includes several subcategories, which specifically describe different types of transactions or business activities. For example, under "Class A events", there are subcategories such as "Personal intra-city remittances", "Courier batch handling fee collection" and "Consumption"; under "Class B events", there are subcategories such as "Repayment and deductions", "Investment" and "Termination", while Class C events only include important events that only occur when customers are lost or only when customers are retained.

[0079] Optionally, based on important events, monitoring rules can be established, and customers can be monitored based on the monitoring rules. When a customer triggers the monitoring rules, a corresponding early warning is implemented for the customer.

[0080] Exemplarily, monitoring rules formed based on important events are as follows:

[0081] Table 4: Cash withdrawal (borrowing) of mobile payment platforms

[0082]

[0083] Monitoring rules: In the past four months, the frequency of cash withdrawal (borrowing) of change on the mobile payment platform was greater than 0.6, and in the past month, cash withdrawal (borrowing) of change on the mobile payment platform did not occur. It can be predicted that the risk of loss of such customers will increase significantly.

[0084] Table 5: Cash withdrawal (loan) from mobile payment platforms

[0085]

[0086] Monitoring rules: If there has been no cash withdrawal (loan) of change on the mobile payment platform in the past three months, it can be predicted that the risk of loss of such customers will increase significantly.

[0087] Figure 5 The flowchart of applying monitoring rules according to an embodiment of the present disclosure is schematically shown.

[0088] like Figure 5As shown in the figure, after successfully mining important events, monitoring rules are summarized based on important events, and these monitoring rules are used to capture abnormal patterns of customer behavior and early signals that may indicate customer churn. By accurately defining and setting these rules, an efficient monitoring platform can be built to monitor customer behavior regularly and automatically. When a customer triggers a monitoring rule, the system will immediately respond and conduct further evaluation. If it is judged that the customer's churn risk is significantly increased, the customer will be identified as a risky customer. This determination process combines multiple advanced technologies such as data mining and machine learning to ensure accuracy and efficiency. Once the list of risky customers is determined, this information will be immediately fed back to the corresponding business department. After receiving this list, the business department will act quickly to develop personalized retention strategies for each risky customer. These strategies may include providing discounts, improving services, strengthening communication, and other methods, aiming to re-stimulate customer interest and loyalty and reduce the risk of customer churn. Therefore, by establishing monitoring rules, banks can monitor customers' financial behavior in real time. Once a customer's behavior triggers specific monitoring rules, the system can respond immediately, thereby achieving real-time monitoring of customer churn risk, enabling banks to detect potential risks in a timely manner and providing time advantages for subsequent early warning and retention work.

[0089] It is understandable that through the analysis of important events, when a customer's transaction behavior shows a pattern similar to that of an important event, timely measures can be taken to reduce the possibility of customer churn.

[0090] Furthermore, generating n3 churn features based on the n1 behavior path features and the n2 event features includes: using a filtering method to filter out n4 primary churn features from the n1 behavior path features and the n2 event features, wherein n4 is less than the sum of n1 and n2 and n4 is greater than n3; and using an embedded method to filter out n3 churn features from the n4 primary churn features.

[0091] Specifically, in the process of feature screening, a two-stage feature selection strategy is adopted. The first stage mainly uses filtering methods, including variance selection method, correlation coefficient method and chi-square test method, which can quickly screen out primary churn features in behavioral path features and event features, remove redundant and irrelevant features, and lay the foundation for subsequent feature selection. Among them, the variance selection method can help remove those features with small variance, that is, those features whose values ​​do not change much, and these features contribute less to the prediction ability of the model. The correlation coefficient rule can measure the strength of the linear relationship between the feature and the target variable, and help screen out features with strong correlation with the target variable. The chi-square test rule is a method for testing the correlation between categorical variables, which is applicable to the customer churn prediction problem in the embodiment of the present disclosure, so as to identify which features have a significant impact on whether the customer churns. In the second stage, embedded methods, including random forests and lightweight gradient boosting machines (LightGBM in English), are used to further select the primary churn features after preliminary screening to obtain churn features. These methods automatically perform feature selection during model training and can more accurately evaluate the importance of features. For example, random forest can evaluate the importance of each feature by calculating the average information gain of each feature at tree splits; while LightGBM can evaluate the importance of features through its built-in feature importance scoring mechanism.

[0092] Exemplarily, four feature selection schemes are designed by combining variance selection, correlation coefficient, chi-square test, and random forest or LightGBM. These schemes have their own advantages and disadvantages in the effect of feature selection. By comparing the prediction results of different schemes, we can select the feature selection scheme that best suits the current data set and problem. Scheme 1 uses variance selection; Scheme 2 uses variance selection, correlation coefficient, and chi-square test; Scheme 3 uses variance selection, correlation coefficient, chi-square test, and random forest; Scheme 4 uses variance selection, correlation coefficient, chi-square test, and LightGBM.

[0093] It can be understood that through the two-stage feature selection strategy, namely the filtering method and the embedded method, it is possible to effectively extract the churn features that reflect the customer churn risk, more comprehensively evaluate the importance of the features, and provide strong support for subsequent model construction or prediction.

[0094] Furthermore, the method of predicting the churn probability of the customer based on the n3 churn features using a pre-built churn prediction model includes: taking the n3 churn features as input and inputting them respectively into k pre-built primary learners to obtain k primary output variables, wherein k is a positive integer greater than or equal to 2; and taking the k primary output variables as input and inputting them into a pre-built secondary learner to obtain the churn probability of the customer, wherein the secondary learner serves as a fusion model to fuse the output variables of the k primary learners.

[0095] Specifically, after obtaining the churn features, the churn features are input into a pre-built churn prediction model, and in the first stage, multiple tree models are used as primary learners. The algorithms corresponding to the multiple tree models may include but are not limited to: a random forest (Random Forest) algorithm that integrates multiple decision trees, which outputs prediction results by majority voting or averaging; an extreme random tree (also called an extreme random forest, Extra Trees) algorithm, which is similar to the random forest algorithm and is more random when selecting split points; an extreme gradient boosting algorithm (XGBoost), which is a gradient boosting-based tree model that can handle missing values ​​and provide efficient training; a lightweight gradient boosting machine (LightGBM) algorithm, which is another gradient boosting-based tree model suitable for large-scale data processing; a classification boosting (CATBoost) algorithm, which is a gradient boosting-based algorithm suitable for processing classification features.

[0096] After the churn features are input into multiple primary learners, multiple primary output variables are obtained. Then, in the second stage, each primary output variable is input into a pre-built secondary learner, and logistic regression is used as the final model fusion algorithm. Specifically, the prediction results of each primary learner in the first stage are used as input variables, and the secondary learner further learns the relationship between these prediction results, thereby optimizing the final prediction output and obtaining the churn probability. The role of logistic regression in this stage is to find the optimal weighted combination of the prediction results of different primary learners, thereby improving the prediction performance of the overall model.

[0097] It is understandable that by using multiple primary learners, the advantages of each model can be fully utilized to capture multiple features and patterns in the data. At the same time, each primary learner may show different deviations in different aspects of the data. Combining the prediction results of these primary learners through secondary learners can reduce the deviation of a single model, improve the accuracy of the overall prediction, and optimize the final prediction output.

[0098] In an embodiment of the present disclosure, the method further includes: based on the n3 churn features, outputting a feature importance list using a pre-built churn prediction model, wherein the feature importance list includes impact values ​​of the n3 churn features, and the impact values ​​of the n3 churn features are used to characterize the importance of the impact of the n3 churn features on the churn probability of the customer.

[0099] Based on the trained churn prediction model, the customer list and feature importance list are finally output as the model results. This list lists the churn probability of each customer in detail, helping banks to more clearly understand which customers have a higher risk of churn. With this list, banks can formulate personalized retention strategies for high-risk customers, thereby improving customer satisfaction and loyalty. At the same time, in order to have a deeper understanding of the working principle of the model and the impact of features on the prediction results, a feature importance list is output, which shows in detail the contribution of each feature in the model prediction, expressed in the feature importance value range [0,1]. The closer the feature importance is to 1, the more important the feature is for judging whether a customer has churned. By analyzing the feature importance list, you can find the features that have the greatest impact on model prediction, help understand the main driving factors of the data, and retain the features with high importance and eliminate the features with low importance during the model training process, thereby simplifying the model and reducing the risk of overfitting. At the same time, understand the relationship between features and target variables, better understand customer needs and behavior patterns, and provide strong support for subsequent customer relationship management and business decisions.

[0100] Specifically, the churn prediction model makes predictions by constructing a series of decision trees, where each tree is optimized based on the residual of the previous tree. In this process, the model evaluates the contribution of each feature to the final prediction result, thereby generating a list of feature importance. There are many ways to calculate feature importance, and the commonly used ones include Gini importance, also known as average reduced purity, which calculates the reduction in impurity (such as Gini index or information gain) of a feature each time a node is split in all trees, and accumulates and normalizes these reductions. Another example is feature importance based on split points, also known as average reduced accuracy, which calculates the contribution of a feature to prediction accuracy when it is used as a split point in all trees, usually based on cross-validation. And, directly count the number of times or depth each feature is selected as a split node through the structure of the tree, and then calculate its importance.

[0101] It can be understood that by generating and analyzing feature importance lists, key features can be identified, feature selection can be performed, and the relationship between data features and target variables can be understood, thereby improving the performance and interpretability of the model.

[0102] Figure 6The structural block diagram of the early warning device for preventing customer churn according to an embodiment of the present disclosure is schematically shown.

[0103] like Figure 6 As shown, the early warning device 600 for preventing customer churn according to this embodiment includes an acquisition module 610 , an analysis module 620 , a generation module 630 , a prediction module 640 and an early warning module 650 .

[0104] The acquisition module 610 is used to acquire m business data stored in the banking business system by the customer with the customer's consent or authorization. In one embodiment, the acquisition module 610 can be used to perform the operation S210 described above, which will not be described in detail here.

[0105] The analysis module 620 is used to perform data analysis on the m business data to obtain n1 behavior path features and n2 event features, wherein the behavior path features are used to characterize the behavior paths of customers who are prone to churn, and the event features are used to characterize the events that occur to customers who are prone to churn, and the storage space occupied by the n1 behavior path features and the n2 event features is less than the storage space occupied by the m business data. In one embodiment, the analysis module 620 can be used to perform the operation S220 described above, which will not be repeated here.

[0106] The generation module 630 is used to generate n3 churn features according to the n1 behavior path features and the n2 event features, wherein the storage space occupied by the n3 churn features is smaller than the storage space occupied by the n1 behavior path features and the n2 event features. In one embodiment, the generation module 630 can be used to perform the operation S230 described above, which will not be described in detail here.

[0107] The prediction module 640 is used to predict the churn probability of the customer based on the n3 churn features using a pre-trained churn prediction model. In one embodiment, the prediction module 640 can be used to perform the operation S240 described above, which will not be described in detail here.

[0108] The early warning module 650 is used to classify the lost customers into risk levels based on the loss probability, and to issue corresponding early warnings to the customers according to the risk levels. In one embodiment, the early warning module 650 can be used to perform the operation S250 described above, which will not be described in detail here.

[0109] In an embodiment of the present disclosure, the prediction module 640 is specifically used to: take the n3 churn features as input, respectively input them into pre-constructed k primary learners to obtain k primary output variables, wherein k is a positive integer greater than or equal to 2; and take the k primary output variables as input, input them into a pre-constructed secondary learner to obtain the churn probability of the customer, wherein the secondary learner serves as a fusion model to fuse the output variables of the k primary learners.

[0110] In an embodiment of the present disclosure, the generation module 630 is specifically used to: use a filtering method to filter out n4 primary churn features from the n1 behavior path features and the n2 event features, wherein n4 is less than the sum of n1 and n2 and n4 is greater than n3; and use an embedded method to filter out n3 churn features from the n4 primary churn features.

[0111] In an embodiment of the present disclosure, the prediction module 640 may also be used to: based on the n3 churn features, output a feature importance list using a pre-trained churn prediction model, wherein the feature importance list includes the impact values ​​of the n3 churn features, and the impact values ​​of the n3 churn features are used to characterize the importance of the impact of the n3 churn features on the churn probability of the customer.

[0112] In an embodiment of the present disclosure, the m business data include the asset data of the customer, and the analysis module 620 is specifically used to: determine the financial behavior of the customer in the current time period based on the asset data of the customer, and divide the customer into at least one of channel dependence, savings, investment and net expenditure types according to the financial behavior to obtain the financial status of the customer; and establish an observation period of a preset time period for the customer based on the financial status of the customer, and analyze the changes in the financial behavior of the customer during the observation period to determine the behavioral path characteristics of the customer.

[0113] In an embodiment of the present disclosure, the analysis module 620 can also be used to: parse the customer's transaction identification based on the customer's asset data, and identify the customer's transaction event based on the transaction identification, wherein the transaction identification is used to determine the transaction information of the customer executing the transaction event; classify the customer's transaction events to form important events; and form the event features based on the important events.

[0114] In an embodiment of the present disclosure, the prediction module 640 can also be used to: obtain the business data of the customer that meets the preset analysis conditions within a historical time period, and perform data analysis to obtain the behavior path characteristics and the event characteristics; use a filtering method and an embedded method to perform feature screening on the behavior path characteristics and the event characteristics to obtain the churn characteristics; input the churn characteristics into multiple pre-built primary learners, and obtain the primary output variables output by each of the primary learners; input each of the primary output variables into a pre-built secondary learner to obtain the churn probability output by the secondary learner; and use a statistical test method to test the churn probability output by the secondary learner, and adjust the parameters of the churn prediction model according to the statistical value calculated by the statistical test method until the statistical distribution of the churn probability is consistent with the statistical distribution of the actual churn situation.

[0115] According to an embodiment of the present disclosure, any multiple modules of the acquisition module 610, the analysis module 620, the generation module 630, the prediction module 640 and the early warning module 650 can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module. According to an embodiment of the present disclosure, at least one of the acquisition module 610, the analysis module 620, the generation module 630, the prediction module 640 and the early warning module 650 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware and firmware or in a suitable combination of any of them. Alternatively, at least one of the acquisition module 610, the analysis module 620, the generation module 630, the prediction module 640 and the warning module 650 may be at least partially implemented as a computer program module, and when the computer program module is executed, the corresponding function may be executed.

[0116] Figure 7 A block diagram schematically shows an electronic device for an early warning method for preventing customer churn according to an embodiment of the present disclosure.

[0117] like Figure 7As shown, the electronic device 700 according to an embodiment of the present invention includes a processor 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage part 708 to a random access memory (RAM) 703. The processor 701 may include, for example, a general-purpose microprocessor (such as a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (for example, an application-specific integrated circuit (ASIC)), etc. The processor 701 may also include an onboard memory for caching purposes. The processor 701 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present invention.

[0118] In RAM 703, various programs and data required for the operation of electronic device 700 are stored. Processor 701, ROM 702 and RAM 703 are connected to each other via bus 704. Processor 701 performs various operations of the method flow according to the embodiment of the present invention by executing the programs in ROM 702 and / or RAM 703. It should be noted that the program can also be stored in one or more memories other than ROM 702 and RAM 703. Processor 701 can also perform various operations of the method flow according to the embodiment of the present invention by executing the programs stored in the one or more memories.

[0119] According to an embodiment of the present invention, the electronic device 700 may further include an input / output (I / O) interface 705, which is also connected to the bus 704. The electronic device 700 may further include one or more of the following components connected to the I / O interface 705: an input portion 706 including a keyboard, a mouse, etc.; an output portion 707 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage portion 708 including a hard disk, etc.; and a communication portion 709 including a network interface card such as a LAN card, a modem, etc. The communication portion 609 performs communication processing via a network such as the Internet. A drive 710 is also connected to the I / O interface 705 as needed. A removable medium 711, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 710 as needed, so that a computer program read therefrom is installed into the storage portion 708 as needed.

[0120] The present invention also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiment; or may exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present invention is implemented.

[0121] According to an embodiment of the present invention, the computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, it may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, the computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, an apparatus or a device. For example, according to an embodiment of the present invention, the computer-readable storage medium may include the ROM 702 and / or RAM 703 described above and / or one or more memories other than ROM 702 and RAM 703.

[0122] The embodiment of the present disclosure also includes a computer program product, which includes a computer program, and the computer program contains program code for executing the method shown in the flowchart. When the computer program product is run in a computer system, the program code is used to enable the computer system to implement the early warning method for preventing customer loss provided by the embodiment of the present disclosure.

[0123] The above functions defined in the system / device of the embodiment of the present disclosure are performed when the computer program is executed by the processor 701. According to the embodiment of the present disclosure, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0124] In one embodiment, the computer program may rely on tangible storage media such as optical storage devices, magnetic storage devices, etc. In another embodiment, the computer program may also be transmitted and distributed in the form of signals on a network medium, and downloaded and installed through the communication part 709, and / or installed from the removable medium 711. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0125] In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 709, and / or installed from the removable medium 711. When the computer program is executed by the processor 701, the above functions defined in the system of the embodiment of the present disclosure are performed. According to the embodiment of the present disclosure, the system, device, means, module, unit, etc. described above can be implemented by a computer program module.

[0126] According to an embodiment of the present disclosure, the program code for executing the computer program provided by the embodiment of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level process and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, Java, C++, python, "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on the remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect through the Internet).

[0127] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0128] It will be appreciated by those skilled in the art that the features described in the various embodiments and / or claims of the present disclosure may be combined and / or combined in a variety of ways, even if such combinations and / or combinations are not explicitly described in the present disclosure. In particular, the features described in the various embodiments and / or claims of the present disclosure may be combined and / or combined in a variety of ways without departing from the spirit and teachings of the present disclosure. All of these combinations and / or combinations fall within the scope of the present disclosure.

[0129] The embodiments of the present disclosure are described above. However, these embodiments are only for illustrative purposes and are not intended to limit the scope of the present disclosure. Although the embodiments are described above separately, this does not mean that the measures in the various embodiments cannot be used in combination to advantage. The scope of the present disclosure is defined by the attached claims and their equivalents. Without departing from the scope of the present disclosure, those skilled in the art may make a variety of substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. An early warning method for preventing customer loss, characterized in that: The method comprises: With the customer's consent or authorization, obtain the customer's m business data stored in the banking business system; Performing data analysis on the m business data to obtain n1 behavior path features and n2 event features, wherein the behavior path features are used to characterize the behavior paths of customers who are prone to churn, and the event features are used to characterize events that occur to customers who are prone to churn, and the storage space occupied by the n1 behavior path features and the n2 event features is smaller than the storage space occupied by the m business data; Generating n3 churn features according to the n1 behavior path features and the n2 event features, wherein the storage space occupied by the n3 churn features is smaller than the storage space occupied by the n1 behavior path features and the n2 event features; Based on the n3 churn features, predicting the churn probability of the customer using a pre-trained churn prediction model; and Based on the churn probability, the customers are classified into risk levels, and corresponding warnings are issued to the customers according to the risk levels.

2. The method according to claim 1, characterized in that: The method of predicting the churn probability of the customer based on the n3 churn features using a pre-trained churn prediction model includes: Input the n3 churn features as inputs into k pre-built primary learners respectively to obtain k primary output variables, where k is a positive integer greater than or equal to 2; and The k primary output variables are input as input into a pre-built secondary learner to obtain the churn probability of the customer, wherein the secondary learner is used as a fusion model to fuse the output variables of the k primary learners.

3. The method according to claim 1 or 2, characterized in that: The step of generating n3 churn features according to the n1 behavior path features and the n2 event features includes: Using a filtering method, n4 primary churn features are screened out from the n1 behavior path features and the n2 event features, wherein n4 is less than the sum of n1 and n2 and n4 is greater than n3; and Using the embedded method, n3 churn features are screened out from the n4 primary churn features.

4. The method according to claim 1, characterized in that: The method further comprises: Based on the n3 churn features, a pre-trained churn prediction model is used to output a feature importance list. The feature importance list includes the influence values ​​of the n3 churn features, and the influence values ​​of the n3 churn features are used to represent the importance of the influence of the n3 churn features on the churn probability of the customer.

5. The method according to claim 1, characterized in that The m business data include the customer's asset data, The performing data analysis on the m business data to obtain n1 behavior path features and n2 event features includes: Determining the financial behavior of the customer in a current time period based on the customer's asset data, and classifying the customer into at least one of a channel-dependent type, a savings type, an investment type, and a net expenditure type according to the financial behavior, so as to obtain the financial status of the customer; and Based on the financial status of the customer, an observation period of a preset time period is established for the customer, and changes in the financial behavior of the customer during the observation period are analyzed to determine the behavioral path characteristics of the customer.

6. The method according to claim 5, characterized in that The performing data analysis on the m business data to obtain n1 behavior path features and n2 event features includes: Based on the asset data of the customer, parsing the transaction identifier of the customer, and identifying the transaction event of the customer according to the transaction identifier, wherein the transaction identifier is used to determine the transaction information of the customer executing the transaction event; Categorizing the transaction events of the customer to form significant events; and The event features are formed according to the important events.

7. The method according to claim 1, characterized in that Training the churn prediction model includes: Acquire the business data of the customer that meets the preset analysis conditions within a historical time period, and perform data analysis to obtain the behavior path characteristics and the event characteristics; Using a filtering method and an embedded method to perform feature screening on the behavior path features and the event features to obtain the churn features; Inputting the churn features into a plurality of pre-built primary learners, and obtaining primary output variables output by each of the primary learners; Inputting each of the primary output variables into a pre-built secondary learner to obtain the churn probability output by the secondary learner; and The churn probability output by the secondary learner is tested using a statistical test method, and the parameters of the churn prediction model are adjusted according to the statistical values ​​calculated by the statistical test method until the statistical distribution of the churn probability conforms to the statistical distribution of the actual churn situation.

8. An early warning device for preventing customer loss, characterized in that: The device comprises: The acquisition module is used to acquire m business data stored by the customer in the banking business system with the customer's consent or authorization; an analysis module, configured to perform data analysis on the m business data to obtain n1 behavior path features and n2 event features, wherein the behavior path features are used to characterize the behavior paths of customers who are prone to churn, and the event features are used to characterize events that occur to customers who are prone to churn, and the storage space occupied by the n1 behavior path features and the n2 event features is smaller than the storage space occupied by the m business data; A generating module, configured to generate n3 churn features according to the n1 behavior path features and the n2 event features, wherein the storage space occupied by the n3 churn features is smaller than the storage space occupied by the n1 behavior path features and the n2 event features; A prediction module, configured to predict the churn probability of the customer using a pre-trained churn prediction model based on the n3 churn features; and The early warning module is used to classify the lost customers into risk levels based on the loss probability, and to issue corresponding early warnings to the customers according to the risk levels.

9. An electronic device, comprising: one or more processors; a storage device for storing one or more computer programs, It is characterized in that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

11. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images