Early warning information generation method and device and electronic equipment

Abstract

The invention discloses a method and device for generating early warning information and electronic equipment, and relates to the field of financial science and technology or other related fields, and the method comprises the steps: collecting customer data of a target customer under the condition of authorization of the target customer, cleaning the customer data, obtaining the target customer data, and storing the target customer data in a database; determining a target feature vector from all the feature vectors based on a preset feature set, processing all the target feature vectors by adopting a preset risk prediction model to obtain a loss probability of the target customer, and when the loss probability is greater than a preset loss threshold value, determining the loss probability of the target customer; and generating early warning information. According to the invention, the technical problem of phases is solved.

Description

Technical Field

[0001] This invention relates to the field of financial technology, and more specifically, to a method, apparatus, and electronic device for generating early warning information. Background Technology

[0002] With the increasing number of banks, competition among them has intensified, making customer churn one of the most significant challenges facing the banking industry. Customer churn refers to the act of a customer ceasing to use an institution's products or services. In the banking industry, customer churn typically manifests as customers closing accounts, transferring funds, reducing transaction frequency, or completely terminating their business relationship with the bank. Customer churn not only directly impacts a bank's revenue but also increases customer acquisition costs (acquiring a new customer typically costs more than five times the cost of retaining an existing one), and can even damage brand reputation. Customer churn can generate negative word-of-mouth, affecting the bank's brand image and market competitiveness.

[0003] With the rapid development of financial technology, customers' expectations for banking services are constantly rising, and banks are facing increasing competitive pressure. This is especially true among younger customer groups, where churn rates are significantly higher than in other age groups. Therefore, effectively predicting customer churn and implementing retention measures has become a pressing issue for banks.

[0004] In related technologies, customer churn prediction and retention methods mainly rely on manual judgment, rule engines, and machine learning models. However, manual judgment is highly subjective and difficult to handle large-scale analysis needs; rule engines judge customer churn risk based on predefined rules (such as account balance below a certain threshold or transaction frequency below a certain threshold), which is difficult to handle complex customer behavior patterns and unstructured data, lacks in-depth analysis of customer behavior, and has low prediction accuracy; machine learning requires a large amount of labeled data, has high model training costs, model results are difficult to interpret, and is highly dependent on data quality and feature engineering, making implementation difficult.

[0005] There is currently no effective solution to the above problems. Summary of the Invention

[0006] This invention provides a method, apparatus, and electronic device for generating early warning information, in order to at least solve certain technical problems.

[0007] According to one aspect of the present invention, a method for generating early warning information is provided, comprising: collecting customer data of a target customer with the authorization of the target customer, wherein the customer data includes at least: transaction data and behavioral data; cleaning the customer data to obtain target customer data, and extracting features from the target customer data to obtain multiple feature vectors; determining a target feature vector from all feature vectors based on a preset feature set, wherein the target feature vector is the feature vector corresponding to each preset feature in the preset feature set; processing all target feature vectors using a preset risk prediction model to obtain the churn probability of the target customer, wherein the preset risk prediction model is a linear model constructed based on a logistic regression algorithm; and generating early warning information when the churn probability is greater than a preset churn threshold.

[0008] Furthermore, before determining the target feature vector from all feature vectors based on the preset feature set, the method further includes: collecting historical data from different data sources under different dimensions, wherein the dimensions include at least one of the following: attributes, transactions, and behaviors; cleaning the historical data to obtain cleaned historical data; extracting features from the cleaned historical data to obtain multiple candidate features; determining the contribution of each candidate feature; identifying the candidate features whose contribution indicators fall within the preset contribution threshold range as preset features, and adding the preset features to the preset feature set.

[0009] Further, the step of cleaning historical data to obtain cleaned historical data includes: performing missing data processing on the historical data to obtain first initial data, wherein missing data processing includes at least one of the following: mean imputation, interpolation, and missing value prediction; identifying outlier data in the first initial data and processing the outlier data to obtain second initial data; and standardizing the second initial data to obtain cleaned historical data.

[0010] Furthermore, the historical data includes: historical customer data for multiple customers, with each customer corresponding to historical churn results. The steps for determining the contribution of each candidate feature include: based on the historical data, determining the feature data corresponding to each candidate feature; for each candidate feature, based on the feature data corresponding to the candidate feature, determining the number of churns and non-churns under the candidate feature, and determining the feature churn ratio based on the number of churns and non-churns; based on the historical churn results, determining the total number of churns and the total number of non-churns, and determining the total churn ratio based on the total number of churns and non-churns; determining the difference between the feature churn ratio and the total churn ratio; performing feature extraction on historical customer data where the historical churn results are churn indicators to obtain the target churn feature vector; performing feature extraction on the feature data to obtain the candidate feature vector of the candidate feature; determining the similarity between the candidate feature vector and the target churn feature vector; and determining the contribution of the candidate feature based on the difference and the similarity.

[0011] Furthermore, before processing all target feature vectors using a pre-defined risk prediction model to obtain the churn probability of target customers, the process includes: performing cluster analysis on historical data to obtain multiple churn categories and setting different churn category labels for each churn category; determining the historical sub-data corresponding to each churn category from the historical data and associating the churn category labels corresponding to the churn categories with the historical sub-data; training the initial risk prediction model using all historical sub-data and the churn category labels associated with the historical sub-data to obtain target model parameters; and adjusting the initial risk prediction model using the target model parameters to obtain the pre-defined risk prediction model.

[0012] Furthermore, after adjusting the initial risk prediction model using the target model parameters to obtain the preset risk prediction model, the process further includes: processing historical data using the preset risk prediction model to obtain prediction results, wherein the prediction results include: prediction tags for each customer; determining preset indicator values ​​for the preset risk prediction model based on the prediction tags and churn category tags; updating the preset feature set when the preset indicator value is less than the preset indicator threshold, and retraining the preset risk prediction model based on the updated preset feature set until the preset indicator value is greater than or equal to the preset indicator threshold.

[0013] Furthermore, the preset risk prediction model includes at least: a linear structure and an activation function. The steps of processing all target feature vectors using the preset risk prediction model to obtain the churn probability of the target customer include: determining the weight corresponding to each target feature vector based on the target model parameters; using a linear structure to linearly combine all target feature vectors and the weights corresponding to each target feature vector based on the logistic regression algorithm to obtain initial values; and processing the initial values ​​using an activation function to obtain the churn probability.

[0014] Furthermore, after generating an early warning message when the churn probability exceeds a preset churn threshold, the process also includes: determining a set of candidate retention strategies based on the churn level contained in the early warning message; conducting comparative tests on each candidate retention strategy in the set of candidate retention strategies to obtain the retention rate of each candidate retention strategy; determining the candidate retention strategy indicated by the maximum retention rate as the target retention strategy, and implementing the target retention strategy on the target customer.

[0015] According to another aspect of the present invention, an apparatus for generating early warning information is also provided, comprising: a collection unit, configured to collect customer data of a target customer with the authorization of the target customer, wherein the customer data includes at least: transaction data and behavioral data; a cleaning unit, configured to clean the customer data to obtain target customer data, and to extract features from the target customer data to obtain multiple feature vectors; a determination unit, configured to determine a target feature vector from all feature vectors based on a preset feature set, wherein the target feature vector is a feature vector corresponding to each preset feature in the preset feature set; a processing unit, configured to process all target feature vectors using a preset risk prediction model to obtain the churn probability of the target customer, wherein the preset risk prediction model is a linear model constructed based on a logistic regression algorithm; and a generation unit, configured to generate early warning information when the churn probability is greater than a preset churn threshold.

[0016] Furthermore, the generation device also includes: a first acquisition module, used to acquire historical data from different data sources under different dimensions before determining the target feature vector from all feature vectors based on a preset feature set, wherein the dimension includes at least one of the following: attribute, transaction, behavior; a first cleaning module, used to clean the historical data to obtain cleaned historical data; a first extraction module, used to extract features from the cleaned historical data to obtain multiple candidate features; a first determination module, used to determine the contribution of each candidate feature; and a second determination module, used to determine the candidate features whose contribution indicators fall within the preset contribution threshold range as preset features and add the preset features to the preset feature set.

[0017] Furthermore, the first cleaning module includes: a first processing submodule, used to perform missing data processing on historical data to obtain first initial data, wherein the missing data processing includes at least one of the following: mean imputation, interpolation, and missing value prediction; a first identification submodule, used to identify abnormal data in the first initial data and process the abnormal data to obtain second initial data; and a second processing submodule, used to perform standardization processing on the second initial data to obtain cleaned historical data.

[0018] Furthermore, the historical data includes: historical customer data for multiple customers, each customer corresponding to historical churn results. The first determination module includes: a first determination submodule, used to determine the feature data corresponding to each candidate feature based on the historical data; a second determination submodule, used to determine the number of churns and non-churns under each candidate feature based on the feature data corresponding to the candidate feature, and to determine the feature churn ratio based on the number of churns and non-churns; a third determination submodule, used to determine the total number of churns and the total number of non-churns based on the historical churn results, and to determine the total churn ratio based on the total number of churns and non-churns; a fourth determination submodule, used to determine the difference between the feature churn ratio and the total churn ratio; a first extraction submodule, used to extract features from historical customer data where the historical churn results indicate churn results, to obtain a target churn feature vector; a second extraction submodule, used to extract features from the feature data, to obtain candidate feature vectors for candidate features; a fifth determination submodule, used to determine the similarity between the candidate feature vector and the target churn feature vector; and a sixth determination submodule, used to determine the contribution of the candidate feature based on the difference and similarity.

[0019] Furthermore, the generation device also includes: a first clustering module, used to perform clustering analysis on historical data before processing all target feature vectors using a preset risk prediction model to obtain the churn probability of target customers, to obtain multiple churn category groups, and to set different churn category labels for each churn category group; a third determination module, used to determine the historical sub-data corresponding to each churn category group from the historical data, and to associate the churn category labels corresponding to the churn category groups with the historical sub-data; a first training module, used to train the initial risk prediction model using all historical sub-data and the churn category labels associated with the historical sub-data to obtain target model parameters; and a first adjustment module, used to adjust the initial risk prediction model using the target model parameters to obtain a preset risk prediction model.

[0020] Furthermore, the generation device also includes: a first processing module, used to process historical data using the preset risk prediction model after adjusting the initial risk prediction model with the target model parameters to obtain a preset risk prediction model, wherein the prediction results include: a prediction label for each customer; a fourth determining module, used to determine the preset index value of the preset risk prediction model based on the prediction label and the churn category label; and a first updating module, used to update the preset feature set when the preset index value is less than the preset index threshold, and retrain the preset risk prediction model based on the updated preset feature set until the preset index value is greater than or equal to the preset index threshold.

[0021] Furthermore, the preset risk prediction model includes at least: a linear structure and an activation function. The processing unit includes: a fifth determining module, used to determine the weights corresponding to each target feature vector based on the target model parameters; a first combining module, used to linearly combine all target feature vectors and the weights corresponding to each target feature vector using a linear structure based on a logistic regression algorithm to obtain initial values; and a second processing module, used to process the initial values ​​using an activation function to obtain the churn probability.

[0022] Furthermore, the generating device also includes: a sixth determining module, used to generate an early warning message when the churn probability is greater than a preset churn threshold, and then determine a set of candidate retention strategies based on the churn level contained in the early warning message; a first testing module, used to perform a comparative test on each candidate retention strategy in the candidate retention strategy set to obtain the retention rate of each candidate retention strategy; and a seventh determining module, used to determine the candidate retention strategy indicated by the maximum retention rate as the target retention strategy and execute the target retention strategy on the target customer.

[0023] According to another aspect of the present invention, a computer program product is also provided, including a non-volatile computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the method for generating any of the above-mentioned warning information.

[0024] According to another aspect of the present invention, an electronic device is also provided, including one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement any of the above-described methods for generating warning information.

[0025] In this invention, with the authorization of the target customer, customer data of the target customer is collected, the customer data is cleaned to obtain target customer data, and features are extracted from the target customer data to obtain multiple feature vectors. Based on a preset feature set, the target feature vector is determined from all feature vectors. A preset risk prediction model is used to process all target feature vectors to obtain the churn probability of the target customer. If the churn probability is greater than a preset churn threshold, an early warning message is generated, thereby solving the relevant technical problem.

[0026] In this invention, multi-dimensional customer data authorized by target customers is first collected. Then, data preprocessing techniques, including data cleaning and feature extraction, are used to obtain a series of feature vectors. Next, a preset feature set is used to filter out features highly correlated with customer churn, constructing a target feature vector set. Then, a preset risk prediction model based on logistic regression is used to calculate the churn probability of target customers in real time. When the churn probability exceeds a preset threshold, an early warning mechanism is automatically triggered, generating warning information so that the bank can take timely retention measures, effectively reducing customer churn rate and improving customer loyalty. This achieves the technical effect of accurately identifying customer churn risk and realizing efficient early warning and prevention of customer churn, thereby solving the technical problem of business losses and waste of technical resources caused by the bank's failure to predict customer churn in advance. Attached Figure Description

[0027] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0028] Figure 1 A hardware structure block diagram of a computer terminal (or mobile device) for implementing a method for generating early warning information is shown.

[0029] Figure 2 This is a flowchart of a method for generating early warning information according to Embodiment 1 of the present invention;

[0030] Figure 3 This is a schematic diagram of a customer churn early warning and retention system based on statistical methods according to Embodiment 1 of the present invention;

[0031] Figure 4 This is a schematic diagram of the churn risk prediction module process according to Embodiment 1 of the present invention;

[0032] Figure 5 This is a schematic diagram of an optional early warning information generation device according to an embodiment of the present invention;

[0033] Figure 6 This is a structural block diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0034] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0035] It should be noted that the terms "first," "second," etc., used in this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0036] It should be noted that all relevant information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, and displayed data) collected and involved in this invention are information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of this data comply with the relevant laws, regulations, and standards of the relevant regions, necessary confidentiality measures have been taken, and it does not violate public order and good morals. Corresponding operation entry points are provided for users to choose to authorize or refuse. For example, this system has an interface with relevant users or organizations. Before obtaining relevant information, a request to obtain the information needs to be sent to the aforementioned user or organization through the interface. The relevant information is obtained only after receiving consent from the aforementioned user or organization. If the user chooses to refuse, the process proceeds to an expert decision-making process.

[0037] This invention, leveraging the rapid development of statistical theory and big data technology, proposes a statistical-based method for customer churn early warning and retention. By integrating statistical methods such as logistic regression, survival analysis, and cluster analysis, it can accurately predict customer churn risk and provide targeted retention strategies, helping banks reduce customer churn rates and improve customer loyalty.

[0038] Current customer churn prediction methods suffer from shortcomings in accuracy and interpretability. Existing customer churn early warning methods (such as rule engines and simple statistical analysis) struggle to handle complex customer behavior patterns, resulting in low accuracy and high false positive rates. This invention introduces statistical methods such as logistic regression, survival analysis, and cluster analysis to more accurately identify risk factors and patterns of customer churn. While traditional machine learning methods (such as random forests) offer high prediction accuracy, their computational complexity is high, requiring substantial computing resources and labeled data, leading to high implementation costs. In contrast, the statistical methods employed in this invention (such as logistic regression and survival analysis) have lower computational complexity and are suitable for real-time processing of large-scale data. Furthermore, many statistical methods (such as cluster analysis) can be directly applied to unlabeled data, reducing data preparation costs. By reducing computational costs, this invention provides banks with a cost-effective customer churn early warning solution.

[0039] This invention integrates statistical methods and big data technology to provide a highly efficient, accurate, and cost-effective customer churn early warning and retention system, helping banks reduce customer churn rates, improve customer loyalty, and minimize business losses. By improving early warning accuracy, reducing computational costs, enhancing interpretability, providing targeted retention strategies, and supporting data-driven decision-making, it offers banks a comprehensive customer churn management solution.

[0040] The present invention will now be described in detail with reference to various embodiments.

[0041] Example 1

[0042] According to an embodiment of this application, an embodiment of a method for generating early warning information is also provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0043] The method embodiment provided in Embodiment 1 of this application can be executed on a mobile terminal, computer terminal, or similar computing device. Figure 1 A hardware block diagram of a computer terminal (or mobile device) for implementing a method for generating early warning information is shown. Figure 1 As shown, computer terminal 10 (or mobile device) may include one or more ( Figure 1The processor 102 (which may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 104 for storing data, and a transmission device 106 for communication functions may also be included. In addition, it may include: a display, a keyboard, a cursor control device, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of a BUS bus), a network interface, a power supply, and / or a camera, wherein the network interface can be connected to wired and / or wireless networks. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0044] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10 (or mobile device). As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0045] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the warning information generation method in this embodiment. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the aforementioned warning information generation method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0046] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0047] The display may be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of the computer terminal 10 (or mobile device).

[0048] Under the aforementioned operating environment, this application provides the following: Figure 2 The method for generating early warning information is shown. Figure 2 This is a flowchart of a method for generating early warning information according to Embodiment 1 of the present invention, as shown below. Figure 2 As shown, the method includes the following steps:

[0049] Step S201: With the authorization of the target customer, collect the target customer's customer data, which includes at least: transaction data and behavioral data.

[0050] In this embodiment of the invention, customer consent must be obtained before acquiring customer data. This can be achieved by signing a data usage agreement or by clearly informing customers of the data usage and privacy protection measures in the bank's terms of service. The collection of customer data covers multiple dimensions, including transaction data, behavioral data, etc., and this data can come from banking systems, CRM (Customer Relationship Management) systems, transaction systems, etc.

[0051] Here, transaction data refers to a customer's financial records with the bank, such as transaction amount, transaction frequency, and transaction time; behavioral data refers to data on customer interactions with the bank, such as login frequency, product usage, and customer service interaction records.

[0052] Step S202: Clean the customer data to obtain target customer data, and extract features from the target customer data to obtain multiple feature vectors.

[0053] In this embodiment of the invention, customer data is cleaned to improve the quality and efficiency of data analysis. Data cleaning includes imputing missing values, removing outliers, and standardizing numerical features. For example, mean imputation is used to handle missing values, and box plots or Z-scores are used to identify outliers. Next, feature extraction is performed on the cleaned target customer data—a process of extracting meaningful information from the raw data. For example, calculating the customer's transaction frequency (number of transactions in the past month) or assessing the rate of change in the customer's account balance. These feature vectors are quantitative representations of specific customer behaviors and are used for model training and prediction.

[0054] Step S203: Based on the preset feature set, determine the target feature vector from all feature vectors, where the target feature vector is the feature vector corresponding to each preset feature in the preset feature set.

[0055] In this embodiment of the invention, a target feature vector is determined from all feature vectors based on a preset feature set, thereby narrowing the feature space and improving model performance and interpretability. The preset feature set is a series of important features determined based on domain knowledge and prior data analysis. Identifying the target feature vector most relevant to customer churn from all feature vectors can be accomplished through statistical testing (such as the chi-square test) or by utilizing methods built into the model (such as L1 regularized logistic regression). This ensures that the model focuses only on those factors that truly influence customer churn decisions.

[0056] Step S204: Process all target feature vectors using a preset risk prediction model to obtain the churn probability of target customers. The preset risk prediction model is a linear model built based on the logistic regression algorithm.

[0057] In this embodiment of the invention, a preset risk prediction model is used to process the selected target feature vector to obtain the churn probability of the target customer. This model is built based on the logistic regression algorithm. Logistic regression is used to handle classification problems, especially suitable for binary classification scenarios, such as whether a customer will churn. The model learns the relationship between the feature vector and the churn status of the target customer, ultimately outputting a value between 0 and 1, i.e., the churn probability. This is a probability estimate of whether a specific customer is likely to leave the bank in the future, which is crucial for guiding customer management and retention strategies.

[0058] Step S205: If the churn probability is greater than the preset churn threshold, generate an early warning message.

[0059] In this embodiment of the invention, if the churn probability calculated by the preset risk prediction model exceeds a preset churn threshold, an early warning message is automatically generated. Here, the churn threshold is a pre-set critical value, representing the highest churn risk the bank is willing to accept. When this threshold is exceeded, it means that the customer has a high probability of churn, and the bank should take immediate action, such as sending retention emails, providing customized services or offers, in order to reverse the churn trend and maintain customer loyalty.

[0060] In summary, the approach involves first collecting multi-dimensional customer data authorized by target customers, then employing data preprocessing techniques, including data cleaning and feature extraction, to obtain a series of feature vectors. Next, a pre-defined feature set is used to filter out features highly correlated with customer churn, constructing a target feature vector set. Subsequently, a pre-defined risk prediction model based on logistic regression is used to calculate the churn probability of target customers in real time. When the churn probability exceeds a pre-defined threshold, an early warning mechanism is automatically triggered, generating warning information so that the bank can take timely retention measures, effectively reducing customer churn rate and improving customer loyalty. This achieves the technical effect of accurately identifying customer churn risk and realizing efficient early warning and prevention of customer churn, thereby solving the technical problem of business losses and wasted technical resources caused by the bank's failure to predict customer churn in advance.

[0061] To improve the accuracy of determining the preset feature set, in the method for generating early warning information provided in Embodiment 1 of this application, before determining the target feature vector from all feature vectors based on the preset feature set, historical data under different dimensions are collected from different data sources, wherein the dimension includes at least one of the following: attribute, transaction, behavior; the historical data is cleaned to obtain cleaned historical data; features are extracted from the cleaned historical data to obtain multiple candidate features; the contribution of each candidate feature is determined; the candidate features whose contribution indicators fall within the preset contribution threshold range are determined as preset features, and the preset features are added to the preset feature set.

[0062] In this embodiment of the invention, historical data is collected from different data sources (such as banking systems, CRM systems, transaction systems, etc.), covering dimensions including at least attributes, transactions, and behaviors. Here, the attribute dimension refers to the customer's basic information; the transaction dimension involves details of the customer's financial activities, such as transaction amount, transaction frequency, and transaction time; the behavior dimension focuses on the customer's interaction patterns, including customer service center query records and online platform activity levels. These diverse data sources ensure the comprehensiveness and accuracy of the information, providing rich material for subsequent data analysis.

[0063] Next, the collected historical data is cleaned, including but not limited to missing value handling (using techniques such as mean imputation and interpolation), outlier detection (using box plots or Z-score methods), and data standardization (such as Z-score standardization and Min-Max standardization) to ensure data quality and eliminate noise interference. Afterward, feature extraction can be performed on the cleaned historical data to generate a series of candidate features. For example, the feature "average deposit change rate over the past three months" can be extracted from transaction data, and "number of times logging into online banking in the past month" can be extracted from behavioral data.

[0064] In this embodiment of the invention, the contribution of each candidate feature is determined using statistical methods or machine learning techniques. For example, statistical methods such as chi-square test and mutual information can be used to measure the correlation between features and customer churn; or machine learning techniques such as L1 regularized logistic regression and feature importance ranking can be used to evaluate the weight of features. The contribution reflects the influence of a feature in the prediction model and is used to distinguish which features are substantially helpful for prediction.

[0065] Next, a preset contribution threshold is determined, which can be adjusted based on business needs and model performance. Only when the contribution of a candidate feature falls within the preset contribution threshold range will that feature be selected and become a preset feature. These preset features are then added to a preset feature set. This preset feature set forms the basis for subsequent risk prediction model training, containing the most influential features for customer churn prediction, thus helping to improve the accuracy and efficiency of the prediction model.

[0066] In this embodiment, firstly, comprehensive and meticulous data collection and cleaning ensures the input quality of the analysis model, avoiding prediction bias caused by data errors or missing data. Secondly, through scientific feature extraction and contribution evaluation, key features with significant predictive power for customer churn risk can be refined, optimizing model performance, reducing model complexity, and achieving faster real-time prediction. Finally, the establishment of a preset feature set provides solid data support for subsequent customer churn probability prediction, ensuring the accuracy of the early warning mechanism.

[0067] To improve the accuracy of cleaning historical data, in the method for generating early warning information provided in Embodiment 1 of this application, the historical data is processed to obtain first initial data, wherein the processing includes at least one of the following: mean imputation, interpolation, and missing value prediction; abnormal data in the first initial data is identified and processed to obtain second initial data; and the second initial data is standardized to obtain cleaned historical data.

[0068] In this embodiment of the invention, missing data processing can be performed on historical data to generate first initial data. Missing data processing includes, but is not limited to, mean imputation, interpolation, and missing value prediction. Mean imputation fills in missing values ​​for a feature by calculating its average value, and is suitable for numerical features. Interpolation methods, particularly linear or polynomial interpolation, can be used in time series data to estimate the value of missing points based on the values ​​of adjacent data points, and are suitable for handling missing values ​​in sequence data such as transaction times or login times. Missing value prediction uses machine learning models (such as regression models) to predict missing values. The model is trained based on non-missing data, and then the model is used to predict missing values. This method has high requirements for data distribution and feature relationships, and is suitable for handling missing values ​​in complex data structures.

[0069] Then, based on the first initial data, outlier data is identified and processed to generate a second initial data set. Here, outlier data refers to observations that significantly deviate from other values ​​in the dataset, possibly due to measurement errors, data entry errors, or extreme events. Outliers are identified by plotting box plots or calculating the Z-score, a statistical method that determines whether an observation is outlier by comparing its deviation from the mean (in standard deviation). Identified outlier data can be processed appropriately, such as deleting it, replacing it with boundary values, or filling it using interpolation methods, to ensure the data's reasonableness and usability.

[0070] Next, the second initial data is standardized to obtain cleaned historical data. Here, data standardization eliminates the influence of different dimensions between features, bringing all features to the same scale, thereby improving the accuracy and efficiency of subsequent analysis. Standardization methods include Z-score standardization and Min-Max standardization. Z-score standardization calculates the deviation of each observation from the mean and divides it by the standard deviation, making the mean of the dataset 0 and the standard deviation 1. Min-Max standardization, on the other hand, scales the data proportionally to fall within a specific interval (e.g., 0 to 1), achieved by subtracting the minimum value and then dividing by the difference between the maximum and minimum values. After standardization, the data is easier to process, resulting in better model training performance.

[0071] For example, the process of cleaning historical data includes: replacing missing values ​​with the sample mean, filling missing values ​​with linear or polynomial interpolation of adjacent data points, or using machine learning (such as regression models) to predict missing values, and using box plots or Z-score methods to identify and handle outliers, as well as standardizing numerical features (such as Z-score standardization or Min-Max standardization) to eliminate the influence of units.

[0072] In this embodiment, firstly, the proper imputation of missing data ensures data integrity and avoids the impact of missing data on the analysis results. Secondly, the identification and handling of outliers ensures the accuracy and reliability of the data, improving the credibility of subsequent analyses. Finally, data standardization provides a unified basis for model training and feature comparison, contributing to the stability and interpretability of the model.

[0073] Optionally, the historical data includes: historical customer data for multiple customers, with each customer corresponding to historical churn results. To improve the accuracy of determining the contribution of each candidate feature, in the method for generating early warning information provided in Embodiment 1 of this application, based on historical data, feature data corresponding to each candidate feature is determined; for each candidate feature, based on the feature data corresponding to the candidate feature, the number of churns and the number of non-churns under the candidate feature are determined, and based on the number of churns and the number of non-churns, the feature churn ratio is determined; based on historical churn results, the total number of churns and the total number of non-churns are determined, and based on the total number of churns and the total number of non-churns, the total churn ratio is determined; the difference between the feature churn ratio and the total churn ratio is determined; feature extraction is performed on historical customer data where historical churn results indicate churn results to obtain a target churn feature vector; feature extraction is performed on the feature data to obtain candidate feature vectors of candidate features; the similarity between the candidate feature vector and the target churn feature vector is determined; based on the difference and the similarity, the contribution of the candidate feature is determined.

[0074] In this embodiment of the invention, each candidate feature is analyzed based on collected historical data to determine the corresponding feature data, including extracting all information related to the candidate feature from transaction data, behavioral data, and attribute data. For example, if the candidate feature is "average deposit change rate over the past three months," then the feature data will include records of the deposit change rate of all customers over the past three months. For each candidate feature, the feature data is further analyzed to determine the number of churned and non-churned customers under the candidate feature, and the feature churn ratio is obtained by calculating the ratio of churned to non-churned customers. For example, for the feature "login frequency," the number of churned and non-churned customers at a specific login frequency level is counted, and then the proportion of churned customers is calculated. Simultaneously, based on historical churn results, i.e., the historical churn status of all customers, the total number of churned and non-churned customers is determined, and the total churn ratio is calculated. The total churn ratio provides an overview of the churn of the entire customer group and serves as a benchmark for evaluating the contribution of different features. Then, the difference between the feature churn ratio and the total churn ratio is determined. The difference reflects the degree of deviation between the churn rate of the customer group under a specific feature and the overall customer group churn rate, and is an important indicator for measuring the correlation between the feature and customer churn.

[0075] In this embodiment of the invention, historical customer data indicating churn results are analyzed to extract target churn feature vectors. These target churn feature vectors contain a characteristic representation of the churned customer group, providing a direct reference for subsequent contribution assessment. Then, feature extraction is further performed on the feature data to obtain candidate feature vectors for candidate features, and the similarity between the candidate feature vectors and the target churn feature vectors is determined. Similarity can be calculated using various methods, such as cosine similarity or Euclidean distance, to measure the correlation between candidate features and churn features. High similarity indicates that the feature is closely related to customer churn and may be a key factor in churn early warning.

[0076] Next, considering both the difference value and similarity, the contribution of the candidate features is determined. Based on the actual situation, the weights corresponding to the difference value and similarity can be determined. After standardizing the difference value and similarity, the contribution of the candidate features is calculated based on the corresponding weights and the standardized difference value and similarity.

[0077] In this embodiment, the extraction and analysis of feature data, especially the statistics of churn and non-churn numbers, provides a quantitative basis for understanding the relationship between features and customer churn. By calculating the difference between the feature churn ratio and the total churn ratio, as well as the similarity between churned customers and candidate features, it is possible to more accurately identify features that have a significant impact on customer churn. This allows for the focus of resources on optimizing the processing of these key features when building the early warning model. In this way, not only is the predictive accuracy of the early warning model improved, but the interpretability of the model results is also ensured, enabling banking personnel to clearly understand which factors may lead to customer churn and thus formulate more targeted retention strategies.

[0078] To improve the accuracy of the preset risk prediction model, in the method for generating early warning information provided in Embodiment 1 of this application, before processing all target feature vectors using the preset risk prediction model to obtain the churn probability of target customers, cluster analysis is performed on historical data to obtain multiple churn categories, and different churn category labels are set for each churn category; historical sub-data corresponding to each churn category is determined from the historical data, and the churn category labels corresponding to the churn category are associated with the historical sub-data; the initial risk prediction model is trained using all historical sub-data and the churn category labels associated with the historical sub-data to obtain target model parameters; the initial risk prediction model is adjusted using the target model parameters to obtain the preset risk prediction model.

[0079] In this embodiment of the invention, based on historical data, clustering analysis algorithms, such as the K-means algorithm, are applied to group customers and identify multiple churn categories with similar behavioral patterns and risk characteristics. This process is based on key dimensions such as customer transaction frequency, account balance changes, and login behavior to reveal different patterns and trends of customer churn. Furthermore, each churn category is assigned a different churn category label, such as "high-risk group," "medium-risk group," and "low-risk group." The labeling serves to visualize the clustering results and is also an important part of model training, helping the model understand the characteristics of different groups.

[0080] In some alternative embodiments, the silhouette coefficient can be used to evaluate the clustering effect during cluster analysis. A silhouette coefficient close to 1 indicates a good clustering effect, while a silhouette coefficient close to -1 indicates a poor clustering effect. If the silhouette coefficient is below a threshold (e.g., 0.5), the number of clusters should be reselected or the algorithm parameters adjusted. Increasing the number of clusters may improve the silhouette coefficient, but it will also increase the computational complexity. Therefore, in practice, a trade-off between complexity and silhouette coefficient is necessary.

[0081] In this embodiment of the invention, historical sub-data corresponding to each churn category is determined from historical data. Historical sub-data is a detailed record for a specific group, containing historical behavior, transaction, and attribute information of customers in that group. Then, the churn category label corresponding to the churn category is associated with the historical sub-data. This ensures that the model training can correlate customer behavior patterns with risk categories, providing "guidance" for model learning. Subsequently, the initial risk prediction model is trained using all historical sub-data and the associated churn category labels. During training, the model attempts to minimize the difference between the predicted results and the actual churn category labels by adjusting its internal parameters. Here, the initial risk prediction model is a statistical model that uses logistic regression to predict the probability of customer churn. Maximum likelihood estimation can be used to optimize the model parameters to obtain optimized target model parameters. These parameters are the result of the model learning from historical data and are crucial for predicting customer churn risk. Finally, the initial risk prediction model is adjusted using the target model parameters to obtain a preset risk prediction model. This model has been optimized based on historical data and can more accurately predict the churn risk of different customer groups, providing valuable early warning information for banks.

[0082] For example, the expression for the preset risk prediction model is: ,in, (i=1,2,…,n) represents customer characteristics, and the coefficients are... (j=0,1,2,…,n) are model parameters, and P (a value between 0 and 1) is the customer churn probability.

[0083] In this embodiment, personalized and refined customer churn early warning is achieved through cluster analysis and model training. Cluster analysis can identify customer groups with similar risk characteristics, rather than treating all customers as a whole. This helps banks more accurately understand the various patterns of customer churn. By assigning labels to each churn category and associating them with historical sub-data, and then using this data to train the model, it is ensured that the model can learn the group-specific risk characteristics, improving the accuracy of predictions. The acquisition of target model parameters and the establishment of preset models further optimize the model's predictive capabilities, enabling the model to assess churn risk more quickly and accurately when faced with new customer data. This helps banks develop effective retention strategies, reduce customer churn, and improve business efficiency and customer loyalty.

[0084] To further improve the accuracy of the preset risk prediction model, in the method for generating early warning information provided in Embodiment 1 of this application, after adjusting the initial risk prediction model using the target model parameters to obtain the preset risk prediction model, the preset risk prediction model is used to process historical data to obtain prediction results. The prediction results include: prediction tags for each customer; determining the preset index value of the preset risk prediction model based on the prediction tags and churn category tags; updating the preset feature set when the preset index value is less than the preset index threshold, and retraining the preset risk prediction model based on the updated preset feature set until the preset index value is greater than or equal to the preset index threshold.

[0085] In this embodiment of the invention, historical data can be used as a validation dataset to verify the predictive performance of the preset risk prediction model. That is, historical data is used as input to the preset risk prediction model, which processes each customer record to predict the likelihood of churn. This process generates prediction results containing a predicted label for each customer, representing the model's classification of customer churn risk, such as "high risk," "medium risk," or "low risk." Then, based on the predicted labels and the churn category labels obtained through cluster analysis, the preset index value of the preset risk prediction model is calculated. Here, the preset index can be various model performance metrics, such as accuracy, recall, and the AUC-ROC curve (Area Under the Receiver Operating Characteristic Curve), used to evaluate the accuracy and robustness of the model's predictions. Accuracy reflects the model's ability to correctly classify customers, while recall focuses on whether the model can find all truly churned customers.

[0086] If the preset indicator value is lower than the preset indicator threshold (a pre-defined standard value used to determine whether the model has reached the expected performance level), it indicates that the current model's predictive performance does not yet meet the requirements, and the preset feature set needs to be updated. The feature set refers to a series of features that the model relies on during prediction, including but not limited to transaction frequency, account balance changes, and customer lifetime length. Feature updates can be performed by adding newly discovered relevant features, optimizing the weights of existing features, or excluding features with less influence. Based on the updated feature set, the preset risk prediction model is retrained until the preset indicator value reaches or exceeds the preset indicator threshold.

[0087] For example, in the initial training, suppose the model's recall rate is only 0.7, which is lower than the preset threshold of 0.8. You can try adding more features that reflect the frequency of customer interactions, such as the number of calls to the customer service hotline and the frequency of online banking logins, and then retrain the model using these updated features until the recall rate reaches or exceeds 0.8.

[0088] In this embodiment, the iterative model training process ensures that the preset risk prediction model can reach the predetermined performance standards, thereby improving the accuracy and timeliness of customer churn warnings. Through fine-tuning of the feature set, the model can better capture the key factors that truly influence customer churn, avoiding false positives and false negatives, and improving the overall effectiveness of the warning system. Furthermore, this performance feedback-based iterative training mechanism can flexibly adapt to data changes and business needs, continuously optimizing the model and ensuring its long-term predictive ability and adaptability.

[0089] Optionally, the preset risk prediction model includes at least: a linear structure and an activation function. In order to improve the accuracy of determining the churn probability, in the method for generating early warning information provided in Embodiment 1 of this application, the weights corresponding to each target feature vector are determined based on the target model parameters; based on the logistic regression algorithm, a linear structure is used to linearly combine all target feature vectors and the weights corresponding to each target feature vector to obtain an initial value; and an activation function is used to process the initial value to obtain the churn probability.

[0090] In this embodiment of the invention, the weight β corresponding to each target feature vector can be determined based on the optimized target model parameters. Here, the weight reflects the importance of each feature in predicting the churn probability. For example, if transaction frequency is highly correlated with churn, its weight will be larger; conversely, if a feature is less correlated with churn, its weight will be relatively smaller. After determining the weights of all features, a linear structure, i.e., a weighted summation method, is used to combine all target feature vectors and their corresponding weights to obtain an initial value S. This process is essentially the core of logistic regression; by multiplying the features by their corresponding weights and summing the results, a comprehensive score is obtained for subsequent churn probability calculations. For example, the expression for the linear structure can be: .

[0091] After obtaining the initial values, an activation function is used to process them, converting the output of the linear combination into a probabilistic form, i.e., the churn probability P. In logistic regression, activation functions, such as the sigmoid function, can be used for this process. The activation function expression is: This function converts the output of a linear combination into a probability value between 0 and 1, intuitively representing the likelihood of customer churn.

[0092] In this embodiment, the accuracy of the preset risk prediction model in customer churn early warning is improved by accurately determining feature weights, using linear combinations and probability transformations. Determining the weights ensures that the model can learn the most critical risk factors based on historical data, making the prediction results closer to reality. The application of linear combinations and activation functions comprehensively considers these factors and transforms them into easily understandable churn probabilities through mathematical methods, providing banks with a clear view of customer risk. Furthermore, by optimizing the algorithm and model parameters, data fluctuations and nonlinear relationships can be effectively handled, enhancing the model's generalization ability and stability.

[0093] To improve the accuracy of determining the target retention strategy, in the method for generating early warning information provided in Embodiment 1 of this application, after generating early warning information when the churn probability is greater than a preset churn threshold, a set of candidate retention strategies is determined based on the churn level included in the early warning information; a comparative test is performed on each candidate retention strategy in the set of candidate retention strategies to obtain the retention rate of each candidate retention strategy; the candidate retention strategy indicated by the maximum retention rate is determined as the target retention strategy, and the target retention strategy is executed on the target customer.

[0094] In this embodiment of the invention, a set of candidate retention strategies can be determined based on the churn level included in the early warning information. The churn level is a customer churn risk level predicted based on statistical methods, and can be "high," "medium," or "low," etc. The set of candidate retention strategies is a series of possible retention measures designed according to the customer's risk level.

[0095] In this embodiment of the invention, a controlled test is performed on each candidate retention strategy in the candidate retention strategy set. The controlled test typically involves setting up one or more experimental groups and a control group. The experimental groups apply different retention strategies, while the control group does not take any retention measures, thereby observing and comparing the effectiveness of the strategies. For example, experimental group A might implement an interest rate discount strategy, experimental group B might implement a points reward strategy, and control group C might not implement any retention measures. After the controlled test, the retention rate of each candidate retention strategy is obtained, which is the proportion of customers who remain active over a period of time after implementing a certain retention strategy. The retention rate calculation can intuitively reflect the effectiveness of the retention strategy and is an important indicator for evaluating whether a strategy is effective. For example, if the retention rate of experimental group A is significantly higher than that of the control group, this indicates that the interest rate discount strategy has a significant positive effect on customer retention. After calculating the retention rates of all candidate retention strategies, the candidate retention strategy indicated by the highest retention rate is determined as the target retention strategy. The target retention strategy is the best-performing strategy in the candidate set, which can effectively reduce customer churn and improve customer retention. Afterwards, target retention strategies are implemented for the target customers (i.e., customers at high risk of churn identified in the early warning information).

[0096] In this embodiment, a closed-loop customer churn early warning and retention system is constructed by combining statistical prediction results with comparative testing of actual retention strategies. The early warning information not only provides predictions of churn risk but also guides the selection and optimization of retention strategies. Comparative testing ensures the effectiveness of the strategies, avoiding wasted resources on ineffective strategies. Strategies that maximize retention rates are applied to high-risk customers, effectively reducing churn rates and improving customer loyalty and the bank's business stability. Furthermore, by continuously optimizing the set of retention strategies and the execution mechanism, the bank can constantly adjust and improve its retention measures to respond to market changes and customer behavior trends, thereby maintaining a competitive edge.

[0097] The following section provides a detailed explanation of another specific implementation method.

[0098] Figure 3 This is a schematic diagram of a customer churn early warning and retention system based on statistical methods according to Embodiment 1 of the present invention, as shown below. Figure 3 As shown, the system can integrate data collected from data sources such as banking systems, CRM systems, and transaction systems (including internal bank data, CRM system data, and transaction data), perform data cleaning and feature extraction, and then input the features into the churn risk prediction module to calculate churn and classify levels. The churn risk prediction module includes: a logistic regression model to predict churn probability; a survival analysis model to estimate retention rate; cluster analysis for risk level classification; and cluster evaluation to assess clustering effectiveness. Following this, based on the output of the churn risk prediction module, a retention strategy recommendation module generates and optimizes strategies. This module includes: strategy generation (different strategies for different clusters); strategy optimization (A / B testing); and dynamic adjustment (real-time adjustment based on results). Furthermore, the system can provide visualizations through a dashboard and generate reports. The dashboard includes heatmaps, bar charts, and Kaplan-Meier curves (survival function estimation). Here, a heatmap is used to display the churn risk distribution of different customer groups. The X-axis represents customer risk level (high, medium, low), the Y-axis represents customer group characteristics (such as transaction frequency, account balance), and the color intensity indicates the probability of churn (the darker the color, the higher the probability of churn). A bar chart is used to display customer retention rates for different strategies. A Kaplan-Meier curve is used to show the trend of customer retention rates over time. The generated report includes: risk points and segmentation results; the churn risk analysis report describes churn risk points, key risk factors, and customer segmentation results. The retention strategy effectiveness report describes the strategy implementation effects and optimization suggestions.

[0099] Figure 4 This is a schematic diagram of the churn risk prediction module process according to Embodiment 1 of the present invention, as shown below. Figure 4 As shown, the input features are: X1: transaction frequency, X2: account balance, and X3: product usage. These features are then input into a logistic regression model (this model undergoes MLE (Maximum Likelihood Estimation) parameter optimization), and the optimized parameters are... The output is the customer churn probability (a value between 0 and 1). After model optimization, model evaluation can be performed to determine if the AUC-ROC is below a threshold. If it is not below the threshold, the model is considered good and churn prediction can be performed. Based on the prediction results, survival analysis is conducted using Kaplan-Meier curves and the Cox proportional hazards model (a regression model). Then, cluster analysis using the K-means algorithm is used to segment customers and evaluate the clustering effect; a higher silhouette function value indicates better results. If the AUC-ROC is below the threshold, features are reselected and parameters are adjusted.

[0100] In this embodiment, a multi-dimensional customer profile is constructed by integrating customer transaction data, behavioral data, and other data. Feature engineering methods are used to extract key features (such as transaction frequency, account balance, and customer lifecycle); cluster analysis (such as the K-means algorithm) is used to categorize customers into high, medium, and low churn risk groups, providing more comprehensive customer insights and supporting accurate churn prediction and personalized retention strategies. Furthermore, statistical methods are comprehensively used to analyze and predict customer lifecycle and churn time points. Clear statistical indicators and visualization results are provided to facilitate understanding and decision-making by business personnel, and lightweight statistical models (such as logistic regression) are used for real-time churn risk prediction, reducing system resource consumption. In addition, personalized retention plans are provided based on customer churn risk, and personalized retention strategies (such as interest rate discounts and dedicated account manager services) are developed for different groups to improve customer loyalty.

[0101] The method for generating early warning information provided in this application first collects multi-dimensional customer data authorized by the target customer, and then uses data preprocessing techniques, including data cleaning and feature extraction, to obtain a series of feature vectors. Next, a preset feature set is used to filter out features highly correlated with customer churn, constructing a target feature vector set. Then, using a preset risk prediction model based on logistic regression, the churn probability of the target customer is calculated in real time. When the churn probability is detected to exceed a preset threshold, an early warning mechanism is automatically triggered to generate early warning information, enabling the bank to take timely retention measures, effectively reducing customer churn rate, and improving customer loyalty. This achieves the technical effect of accurately identifying customer churn risk and realizing efficient early warning and prevention of customer churn, thereby solving the technical problem of business losses and waste of technical resources caused by the bank's failure to predict customer churn in advance.

[0102] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0103] Example 2

[0104] This application also provides an apparatus for generating early warning information. It should be noted that the apparatus for generating early warning information in this application can be used to execute the method for generating early warning information provided in this application. The apparatus for generating early warning information provided in this application will be described below.

[0105] According to an embodiment of this application, an apparatus for implementing the above-described method for generating early warning information is also provided. Figure 5 This is a schematic diagram of an optional early warning information generation device according to an embodiment of the present invention, such as... Figure 5 As shown, the generating device may include: a collection unit 50, a cleaning unit 51, a determination unit 52, a processing unit 53, and a generating unit 54.

[0106] The collection unit 50 is used to collect customer data of the target customer with the authorization of the target customer. The customer data includes at least: transaction data and behavioral data.

[0107] The cleaning unit 51 is used to clean customer data to obtain target customer data, and to extract features from the target customer data to obtain multiple feature vectors.

[0108] The determining unit 52 is used to determine a target feature vector from all feature vectors based on a preset feature set, wherein the target feature vector is the feature vector corresponding to each preset feature in the preset feature set;

[0109] Processing unit 53 is used to process all target feature vectors using a preset risk prediction model to obtain the churn probability of target customers. The preset risk prediction model is a linear model built based on the logistic regression algorithm.

[0110] The generation unit 54 is used to generate early warning information when the churn probability is greater than a preset churn threshold.

[0111] The early warning information generation device provided in this application embodiment can first collect multi-dimensional customer data authorized by target customers, and then use data preprocessing techniques, including data cleaning and feature extraction, to obtain a series of feature vectors. Then, a preset feature set is used to filter out features highly correlated with customer churn, constructing a target feature vector set. Subsequently, a preset risk prediction model based on logistic regression algorithm is used to calculate the churn probability of target customers in real time. When the churn probability is detected to exceed a preset threshold, an early warning mechanism is automatically triggered to generate early warning information, enabling banks to take timely retention measures, effectively reducing customer churn rate, and improving customer loyalty. This achieves the technical effect of accurately identifying customer churn risk and realizing efficient early warning and prevention of customer churn, thereby solving the technical problem of business losses and waste of technical resources caused by banks' failure to predict customer churn in advance.

[0112] Optionally, the generation device further includes: a first acquisition module, used to acquire historical data from different data sources under different dimensions before determining the target feature vector from all feature vectors based on a preset feature set, wherein the dimension includes at least one of the following: attribute, transaction, behavior; a first cleaning module, used to clean the historical data to obtain cleaned historical data; a first extraction module, used to extract features from the cleaned historical data to obtain multiple candidate features; a first determination module, used to determine the contribution of each candidate feature; and a second determination module, used to determine the candidate features whose contribution indicators fall within the preset contribution threshold range as preset features and add the preset features to the preset feature set.

[0113] Optionally, the first cleaning module includes: a first processing submodule for performing missing data processing on historical data to obtain first initial data, wherein the missing data processing includes at least one of the following: mean imputation, interpolation, and missing value prediction; a first identification submodule for identifying abnormal data in the first initial data and processing the abnormal data to obtain second initial data; and a second processing submodule for performing standardization processing on the second initial data to obtain cleaned historical data.

[0114] Optionally, the historical data includes: historical customer data for multiple customers, each customer corresponding to historical churn results. The first determining module includes: a first determining submodule, used to determine the feature data corresponding to each candidate feature based on the historical data; a second determining submodule, used to determine the number of churns and non-churns under each candidate feature based on the feature data corresponding to the candidate feature, and to determine the feature churn ratio based on the number of churns and non-churns; a third determining submodule, used to determine the total number of churns and the total number of non-churns based on the historical churn results, and to determine the total churn ratio based on the total number of churns and non-churns; a fourth determining submodule, used to determine the difference between the feature churn ratio and the total churn ratio; a first extraction submodule, used to extract features from historical customer data where the historical churn results indicate churn results, to obtain a target churn feature vector; a second extraction submodule, used to extract features from the feature data, to obtain candidate feature vectors for candidate features; a fifth determining submodule, used to determine the similarity between the candidate feature vector and the target churn feature vector; and a sixth determining submodule, used to determine the contribution of the candidate feature based on the difference and similarity.

[0115] Optionally, the generation device further includes: a first clustering module, used to perform clustering analysis on historical data before processing all target feature vectors using a preset risk prediction model to obtain the churn probability of target customers, to obtain multiple churn category groups, and to set different churn category labels for each churn category group; a third determination module, used to determine the historical sub-data corresponding to each churn category group from the historical data, and to associate the churn category labels corresponding to the churn category groups with the historical sub-data; a first training module, used to train the initial risk prediction model using all historical sub-data and the churn category labels associated with the historical sub-data to obtain target model parameters; and a first adjustment module, used to adjust the initial risk prediction model using the target model parameters to obtain a preset risk prediction model.

[0116] Optionally, the generation device further includes: a first processing module, used to process historical data using the preset risk prediction model after adjusting the initial risk prediction model with the target model parameters to obtain a preset risk prediction model, wherein the prediction results include: a prediction label for each customer; a fourth determining module, used to determine the preset index value of the preset risk prediction model based on the prediction label and the churn category label; and a first updating module, used to update the preset feature set when the preset index value is less than the preset index threshold, and retrain the preset risk prediction model based on the updated preset feature set until the preset index value is greater than or equal to the preset index threshold.

[0117] Optionally, the preset risk prediction model includes at least: a linear structure and an activation function. The processing unit includes: a fifth determining module, used to determine the weights corresponding to each target feature vector based on the target model parameters; a first combining module, used to linearly combine all target feature vectors and the weights corresponding to each target feature vector using a linear structure based on a logistic regression algorithm to obtain initial values; and a second processing module, used to process the initial values ​​using an activation function to obtain the churn probability.

[0118] Optionally, the generating device further includes: a sixth determining module, used to generate an early warning message when the churn probability is greater than a preset churn threshold, and then determine a set of candidate retention strategies based on the churn level contained in the early warning message; a first testing module, used to perform a comparative test on each candidate retention strategy in the set of candidate retention strategies to obtain the retention rate of each candidate retention strategy; and a seventh determining module, used to determine the candidate retention strategy indicated by the maximum retention rate as the target retention strategy and execute the target retention strategy on the target customer.

[0119] The aforementioned generating device may also include a processor and a memory. The aforementioned acquisition unit 50, cleaning unit 51, determination unit 52, processing unit 53, generating unit 54, etc., are all stored in the memory as program units, and the processor executes the aforementioned program units stored in the memory to realize the corresponding functions.

[0120] The aforementioned processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured, and by adjusting kernel parameters, a warning message can be generated when the churn probability exceeds a preset churn threshold.

[0121] The aforementioned memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0122] It should be noted that the acquisition unit 50, cleaning unit 51, determination unit 52, processing unit 53, and generation unit 54 mentioned above correspond to steps S201 to S205 in Embodiment 1. The instances and application scenarios implemented by the above units and corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1. It should be noted that the above units can be hardware or software components stored in memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, ..., 102n). The above units can also be part of a device and run in the computer terminal 10 provided in Embodiment 1.

[0123] Example 3

[0124] Embodiments of this application may provide an electronic device. Figure 6 This is a structural block diagram of an electronic device according to an embodiment of the present invention. Figure 6 As shown, the electronic device may include: one or more ( Figure 6 (Only one is shown) Processor 602, memory 604, memory controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0125] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the warning information generation method and apparatus in this application embodiment. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby realizing the aforementioned warning information generation method. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0126] The processor can access information and applications stored in the memory via a transmission device to perform the following steps: With the authorization of the target customer, collect customer data from the target customer, wherein the customer data includes at least transaction data and behavioral data; clean the customer data to obtain target customer data, and extract features from the target customer data to obtain multiple feature vectors; based on a preset feature set, determine a target feature vector from all feature vectors, wherein the target feature vector is the feature vector corresponding to each preset feature in the preset feature set; process all target feature vectors using a preset risk prediction model to obtain the churn probability of the target customer, wherein the preset risk prediction model is a linear model constructed based on a logistic regression algorithm; generate an early warning message if the churn probability is greater than a preset churn threshold.

[0127] The processor can invoke information and applications stored in the memory through the transmission device to perform the following steps: collect historical data from different data sources under different dimensions, wherein the dimensions include at least one of the following: attributes, transactions, behaviors; clean the historical data to obtain cleaned historical data; extract features from the cleaned historical data to obtain multiple candidate features; determine the contribution of each candidate feature; determine the candidate features whose contribution indicators fall within the preset contribution threshold range as preset features, and add the preset features to the preset feature set.

[0128] The processor can invoke information and application programs stored in the memory through the transmission device to perform the following steps: perform missing data processing on historical data to obtain first initial data, wherein missing data processing includes at least one of the following: mean imputation, interpolation, and missing value prediction; identify abnormal data in the first initial data and process the abnormal data to obtain second initial data; perform standardization processing on the second initial data to obtain cleaned historical data.

[0129] The processor can access information and applications stored in memory via a transmission device to perform the following steps: Based on historical data, determine the feature data corresponding to each candidate feature; for each candidate feature, based on the feature data corresponding to the candidate feature, determine the number of churned and non-churned users under the candidate feature, and determine the feature churn ratio based on the number of churned and non-churned users; based on historical churn results, determine the total number of churned and non-churned users, and determine the total churn ratio based on the total number of churned and non-churned users; determine the difference between the feature churn ratio and the total churn ratio; extract features from historical customer data where historical churn results indicate churn results to obtain a target churn feature vector; extract features from the feature data to obtain candidate feature vectors for candidate features; determine the similarity between the candidate feature vector and the target churn feature vector; and determine the contribution of the candidate feature based on the difference and similarity.

[0130] The processor can access the information and application programs stored in the memory via the transmission device to perform the following steps: perform cluster analysis on historical data to obtain multiple churn categories, and set different churn category labels for each churn category; determine the historical sub-data corresponding to each churn category from the historical data, and associate the churn category labels corresponding to the churn categories with the historical sub-data; train the initial risk prediction model using all historical sub-data and the churn category labels associated with the historical sub-data to obtain the target model parameters; adjust the initial risk prediction model using the target model parameters to obtain the preset risk prediction model.

[0131] The processor can access the information and application programs stored in the memory via the transmission device to perform the following steps: process historical data using a preset risk prediction model to obtain prediction results, wherein the prediction results include: prediction tags for each customer; determine the preset index value of the preset risk prediction model based on the prediction tags and churn category tags; update the preset feature set if the preset index value is less than the preset index threshold, and retrain the preset risk prediction model based on the updated preset feature set until the preset index value is greater than or equal to the preset index threshold.

[0132] The processor can access the information and application programs stored in the memory via the transmission device to perform the following steps: determining the weights corresponding to each target feature vector based on the target model parameters; using a linear structure based on the logistic regression algorithm to linearly combine all target feature vectors and the weights corresponding to each target feature vector to obtain initial values; and processing the initial values ​​using an activation function to obtain the churn probability.

[0133] The processor can access the information and application stored in the memory via the transmission device to perform the following steps: determine a set of candidate retention strategies based on the churn level contained in the warning information; conduct comparative tests on each candidate retention strategy in the set of candidate retention strategies to obtain the retention rate of each candidate retention strategy; determine the candidate retention strategy indicated by the maximum retention rate as the target retention strategy, and execute the target retention strategy on the target customer.

[0134] This application provides a scheme for generating early warning information. It involves first collecting multi-dimensional customer data authorized by target customers, then employing data preprocessing techniques, including data cleaning and feature extraction, to obtain a series of feature vectors. Next, a preset feature set is used to filter features highly correlated with customer churn, constructing a target feature vector set. Then, a preset risk prediction model based on logistic regression is used to calculate the churn probability of target customers in real time. When the churn probability exceeds a preset threshold, an early warning mechanism is automatically triggered, generating early warning information so that the bank can take timely retention measures, effectively reducing customer churn rate and improving customer loyalty. This achieves the technical effect of accurately identifying customer churn risk and realizing efficient early warning and prevention of customer churn, thereby solving the technical problem of business losses and waste of technical resources caused by the bank's failure to predict customer churn in advance.

[0135] Those skilled in the art will understand that Figure 6 The structure shown is for illustrative purposes only. Electronic devices can also be terminal devices such as smartphones, tablets, PDAs, and mobile internet devices (MIDs). Figure 6 This does not limit the structure of the aforementioned electronic device. For example, electronic devices may also include components that are more... Figure 6 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 6 The different configurations shown.

[0136] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0137] Example 4

[0138] Embodiments of this application also provide a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the method for generating early warning information provided in Embodiment 1.

[0139] Optionally, in this embodiment, the storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals.

[0140] This application also provides a computer program product, which, when executed on a data processing device, is suitable for performing the steps of a method for generating early warning information.

[0141] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0142] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0143] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0144] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0145] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0146] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0147] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for generating early warning information, characterized in that, include: With the authorization of the target customer, customer data of the target customer is collected, wherein the customer data includes at least: transaction data and behavioral data; The customer data is cleaned to obtain target customer data, and features are extracted from the target customer data to obtain multiple feature vectors; Based on a preset feature set, a target feature vector is determined from all the feature vectors, wherein the target feature vector is the feature vector corresponding to each preset feature in the preset feature set; All the target feature vectors are processed using a preset risk prediction model to obtain the churn probability of the target customer, wherein the preset risk prediction model is a linear model constructed based on the logistic regression algorithm; If the churn probability is greater than a preset churn threshold, an early warning message is generated.

2. The generation method according to claim 1, characterized in that, Before determining the target feature vector from all the feature vectors based on a preset feature set, the process further includes: Historical data from different data sources is collected from different dimensions, wherein the dimensions include at least one of the following: attributes, transactions, and behaviors; The historical data is cleaned to obtain the cleaned historical data; Feature extraction is performed on the cleaned historical data to obtain multiple candidate features; Determine the contribution of each candidate feature; The candidate features of the contribution indication that fall within the preset contribution threshold range are identified as the preset features, and the preset features are added to the preset feature set.

3. The generation method according to claim 2, characterized in that, The step of cleaning the historical data to obtain the cleaned historical data includes: The historical data is processed to obtain the first initial data, wherein the missing data processing includes at least one of the following: mean imputation, interpolation, and missing value prediction; Identify abnormal data in the first initial data and process the abnormal data to obtain the second initial data; The second initial data is standardized to obtain the cleaned historical data.

4. The generation method according to claim 2, characterized in that, The historical data includes: historical customer data for multiple customers, with each customer corresponding to historical churn results. The step of determining the contribution of each candidate feature includes: Based on the historical data, the feature data corresponding to each candidate feature is determined; For each candidate feature, based on the feature data corresponding to the candidate feature, determine the number of lost and non-lost items under the candidate feature, and based on the number of lost and non-lost items, determine the feature loss ratio; Based on the historical churn results, the total number of churned individuals and the total number of non-churned individuals are determined, and based on the total number of churned individuals and the total number of non-churned individuals, the total churn ratio is determined; Determine the difference between the characteristic churn rate and the total churn rate; Feature extraction is performed on the historical customer data that indicates the historical churn result to obtain the target churn feature vector; Feature extraction is performed on the feature data to obtain the candidate feature vector of the candidate feature; Determine the similarity between the candidate feature vector and the target loss feature vector; The contribution of the candidate feature is determined based on the difference value and the similarity.

5. The generation method according to claim 2, characterized in that, Before processing all the target feature vectors using a preset risk prediction model to obtain the churn probability of the target customer, the process further includes: Cluster analysis is performed on the historical data to obtain multiple churn categories, and different churn category labels are assigned to each of the churn categories. Determine the historical sub-data corresponding to each churn category group from the historical data, and associate the churn category label corresponding to the churn category group with the historical sub-data; The initial risk prediction model is trained using all the historical sub-data and the churn category labels associated with the historical sub-data to obtain the target model parameters; The initial risk prediction model is adjusted using the target model parameters to obtain the preset risk prediction model.

6. The generation method according to claim 5, characterized in that, After adjusting the initial risk prediction model using the target model parameters to obtain the preset risk prediction model, the method further includes: The historical data is processed using the preset risk prediction model to obtain prediction results, wherein the prediction results include: prediction tags for each customer; Based on the predicted label and the churn category label, determine the preset index value of the preset risk prediction model; If the preset indicator value is less than the preset indicator threshold, the preset feature set is updated, and the preset risk prediction model is retrained based on the updated preset feature set until the preset indicator value is greater than or equal to the preset indicator threshold.

7. The generation method according to claim 6, characterized in that, The preset risk prediction model includes at least: a linear structure and an activation function. The step of processing all the target feature vectors using the preset risk prediction model to obtain the churn probability of the target customer includes: Based on the target model parameters, determine the weight corresponding to each target feature vector; Based on the logistic regression algorithm, the initial value is obtained by linearly combining all the target feature vectors and the weights corresponding to each target feature vector using the linear structure. The initial value is processed using the activation function to obtain the churn probability.

8. The generation method according to claim 1, characterized in that, After generating an early warning message when the churn probability exceeds a preset churn threshold, the system further includes: Based on the churn level included in the warning information, a set of candidate retention strategies is determined; Each candidate retention strategy in the candidate retention strategy set is tested against the others to obtain the retention rate of each candidate retention strategy; The candidate retention strategy indicated by the maximum retention rate is determined as the target retention strategy, and the target retention strategy is executed on the target customer.

9. A device for generating early warning information, characterized in that, include: The data collection unit is used to collect customer data of the target customer with the authorization of the target customer, wherein the customer data includes at least: transaction data and behavioral data; The cleaning unit is used to clean the customer data to obtain target customer data, and to extract features from the target customer data to obtain multiple feature vectors. A determining unit is configured to determine a target feature vector from all the feature vectors based on a preset feature set, wherein the target feature vector is a feature vector corresponding to each preset feature in the preset feature set; The processing unit is used to process all the target feature vectors using a preset risk prediction model to obtain the churn probability of the target customer, wherein the preset risk prediction model is a linear model constructed based on the logistic regression algorithm; The generation unit is used to generate early warning information when the churn probability is greater than a preset churn threshold.

10. An electronic device, characterized in that, It includes one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method for generating warning information according to any one of claims 1 to 8.

Images