Delinquency stage prediction method and apparatus, electronic device, and storage medium
Patent Information
- Application Number
- PCT/CN2025/096101
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-02-26
- Filing Date
- 2025-05-20
- Publication Date
- 2026-09-03
Smart Images
Figure CN2025096101_03092026_PF_FP_ABST
Abstract
Description
Methods, devices, electronic equipment and storage media for predicting overdue periods
[0001] Cross-references to related applications
[0002] This application claims priority to Chinese Patent Application No. 2025102174788, filed on February 26, 2025, entitled “Method, Apparatus, Electronic Device and Storage Medium for Predicting Overdue Stages”, the entire contents of which are incorporated herein by reference. Technical Field
[0003] This invention relates to the field of computer technology, and more specifically, to a method, apparatus, electronic device, and storage medium for predicting overdue periods. Background Technology
[0004] If a user fails to repay a loan within the stipulated period, it is considered overdue. For users in an overdue state, collection efforts will be initiated to urge them to repay as soon as possible. Therefore, in financial collection scenarios, predicting whether a user will default on their loan repayments is crucial.
[0005] In existing technologies, users' daily data is typically collected to predict their delinquency rates. This method requires processing a large amount of data, which is not only inefficient but also makes it difficult to guarantee the accuracy of the predictions.
[0006] Application content
[0007] In view of this, the purpose of the present invention is to provide a method, apparatus, electronic device and storage medium for predicting overdue periods, so as to improve prediction efficiency and enhance prediction accuracy.
[0008] To achieve the above objectives, the technical solutions adopted in the embodiments of the present invention are as follows:
[0009] In a first aspect, the present invention provides a method for predicting overdue periods, the method comprising:
[0010] Obtain the overdue characteristics of the user in each historical overdue period; the historical overdue period includes multiple consecutive days;
[0011] The overdue features within each of the aforementioned historical overdue stages are input into a pre-trained overdue stage prediction model; the overdue stage prediction model employs a bidirectional LSTM network.
[0012] The overdue stage prediction model predicts the overdue stage at which a user will be overdue next.
[0013] In an optional implementation, inputting the overdue features within each of the historical overdue stages into a pre-trained overdue stage prediction model includes:
[0014] The overdue features within each of the historical overdue stages are input into the forward LSTM network of the overdue stage prediction model in a first order, and the overdue features within each of the historical overdue stages are input into the inverse LSTM network of the overdue stage prediction model in a second order; the first order and the second order are determined according to the order of overdue periods corresponding to each of the historical overdue stages, and the first order and the second order are reversed.
[0015] In an optional implementation, the overdue stage prediction model is trained in the following manner:
[0016] A training dataset is generated based on the overdue characteristics of different users across multiple historical overdue periods.
[0017] The overdue stage prediction model based on the bidirectional LSTM network is trained using the training dataset to obtain the trained overdue stage prediction model.
[0018] In an optional implementation, the method further includes:
[0019] After the overdue stage prediction model is trained, the LRP algorithm is used to evaluate the importance of each overdue feature.
[0020] In an optional implementation, the use of the LRP algorithm to assess the importance of each overdue feature includes:
[0021] Based on the correlation of the current layer of the overdue stage prediction model, the correlation calculation results of the current layer to the previous layer of the forward LSTM network in the overdue stage prediction model and the correlation calculation results of the current layer to the previous layer of the reverse LSTM network in the overdue stage prediction model are calculated respectively.
[0022] The correlation calculation results from the current layer to the previous layer of the forward LSTM network and the correlation calculation results from the current layer to the previous layer of the reverse LSTM network are weighted and calculated to obtain the correlation of the current layer to the previous layer of the overdue stage prediction model.
[0023] Through iterative calculations layer by layer, the correlation of each overdue feature is finally output through the overdue stage prediction model; the correlation of each overdue feature in the input layer characterizes the importance of each overdue feature.
[0024] In an optional implementation, the formula for calculating the correlation between the current layer and the previous layer of the forward LSTM network is as follows:
[0025] Where i represents the previous layer, j represents the current layer, and R i←j R represents the correlation calculation result from the current layer to the previous layer of the forward LSTM network.j The z represents the correlation of the current layer of the overdue stage prediction model. i w represents the output of the i-th layer of the forward LSTM network. ij This represents the weight of the connection between the i-th and j-th layers of the forward LSTM network;
[0026] The formula for calculating the correlation between the current layer and the previous layer of the reverse LSTM network is as follows:
[0027] Among them, R′ i←j This represents the correlation calculation result from the current layer to the previous layer of the inverse LSTM network, z′. i w′ represents the output of the i-th layer of the inverse LSTM network. ij This represents the weight of the connection between the i-th and j-th layers of the reverse LSTM network.
[0028] In a second aspect, the present invention provides an overdue stage prediction device, the device comprising:
[0029] The acquisition module is used to acquire the overdue characteristics of a user in each historical overdue period; the historical overdue period includes multiple consecutive days;
[0030] The input module is used to input the overdue features of each of the historical overdue stages into a pre-trained overdue stage prediction model; the overdue stage prediction model adopts a bidirectional LSTM network.
[0031] The prediction module is used to predict the next overdue stage of a user through the overdue stage prediction model.
[0032] In an optional implementation, the input module is used to input the overdue features in each of the historical overdue stages into the forward LSTM network of the overdue stage prediction model in a first order, and to input the overdue features in each of the historical overdue stages into the inverse LSTM network of the overdue stage prediction model in a second order; the first order and the second order are determined according to the overdue order corresponding to each of the historical overdue stages, and the first order and the second order are opposite.
[0033] Thirdly, the present invention provides an electronic device including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the overdue stage prediction method as described in any of the foregoing embodiments.
[0034] Fourthly, the present invention provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the overdue stage prediction method as described in any of the foregoing embodiments.
[0035] The present invention provides a method, apparatus, electronic device, and storage medium for predicting overdue periods. The method includes: acquiring overdue features within each historical overdue period of a user; each historical overdue period includes multiple consecutive days; inputting the overdue features within each historical overdue period into a pre-trained overdue period prediction model; the overdue period prediction model employs a bidirectional LSTM network; and predicting the overdue period in which the user will be overdue next. By predicting the user's overdue status based on the statistical features of overdue periods, compared to collecting daily user data, the amount of data processing is greatly reduced, and the prediction efficiency is improved. Furthermore, the use of a bidirectional LSTM network can comprehensively capture the user's overdue features from two directions, enhancing prediction accuracy.
[0036] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description
[0037] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0038] Figure 1 shows a flowchart of a method for predicting overdue periods provided in an embodiment of the present invention;
[0039] Figure 2 shows another flowchart of the overdue stage prediction method provided in an embodiment of the present invention;
[0040] Figure 3 shows a functional block diagram of the overdue stage prediction device provided in an embodiment of the present invention;
[0041] Figure 4 shows another functional block diagram of the overdue stage prediction device provided in an embodiment of the present invention;
[0042] Figure 5 shows a block diagram of an electronic device provided in an embodiment of the present invention.
[0043] Icons: 100 - Electronic device; 110 - Memory; 120 - Processor; 130 - Communication module; 600 - Overdue stage prediction device; 610 - Acquisition module; 620 - Input module; 630 - Prediction module; 640 - Calculation module. Detailed Implementation
[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.
[0045] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.
[0046] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0047] Traditional machine learning algorithms or time series models typically use daily data to predict user delinquency, requiring the processing of large amounts of data. This not only results in low prediction efficiency but also makes it difficult to guarantee prediction accuracy. Furthermore, the models suffer from poor interpretability; for example, existing models struggle to provide a clear explanation for why a particular user defaults, making it difficult to provide accurate and interpretable evidence for optimizing subsequent collection strategies.
[0048] Based on this, embodiments of the present invention provide a method, apparatus, electronic device, and storage medium for predicting overdue stages. By predicting user overdue status according to the statistical characteristics of overdue stages, compared to collecting daily user data, the amount of data processing is greatly reduced, and the prediction efficiency is improved. Furthermore, by employing a bidirectional LSTM network, the overdue characteristics of users can be captured comprehensively from two directions, enhancing prediction accuracy. In addition, by introducing the LRP (Layer-wise Relevance Propagation) algorithm, the importance of each overdue feature in the model input layer is evaluated, clearly identifying which overdue features have the greatest impact on the prediction results, thereby improving the model's interpretability and providing a more accurate and interpretable basis for subsequent collection strategy optimization.
[0049] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
[0050] Please refer to Figure 1, which is a flowchart illustrating a method for predicting overdue stages according to an embodiment of the present invention. It should be noted that the method for predicting overdue stages of the present invention is not limited to the specific order shown in Figure 1 and below. It should be understood that in other embodiments, the order of some steps in the method for predicting overdue stages of the present invention can be interchanged according to actual needs, or some steps can be omitted or deleted. This method for predicting overdue stages can be applied to electronic devices such as laptops, tablets, PCs (Personal Computers), and servers. The specific flow shown in Figure 1 will be described in detail below.
[0051] Step S101: Obtain the overdue characteristics of the user in each historical overdue period; the historical overdue period includes multiple consecutive days.
[0052] In this embodiment, the overdue period can be measured in months or quarters, depending on the actual business scenario. For example, a user's historical overdue periods may include overdue for one month, two months, etc., denoted as m1 and m2 respectively.
[0053] For each historical overdue period of a user, the overdue characteristics of the user in that historical overdue period are statistically analyzed. These overdue characteristics may include: number of overdue days, overdue amount, outstanding amount, loan app usage click statistics, car GPS speed, distance statistics, etc.
[0054] Step S102: Input the overdue features of each historical overdue stage into the pre-trained overdue stage prediction model; the overdue stage prediction model adopts a bidirectional LSTM network.
[0055] In this embodiment, the LSTM (Long Short-Term Memory) network is a special neural network structure that excels at processing time-series data. This embodiment employs a bidirectional LSTM network, which can simultaneously consider information from both forward and reverse time, thereby gaining a more comprehensive understanding of user behavior patterns and improving the accuracy of predictions for future delinquency periods.
[0056] Step S103: Predict the overdue stage of the user's next overdue payment using the overdue stage prediction model.
[0057] The overdue phase indicates the duration of the overdue period.
[0058] Optionally, when performing statistics, they can be based on a preset overdue unit. In some embodiments, the preset overdue unit can be: weekly overdue, monthly overdue, quarterly overdue, annual overdue, etc., which is not limited here.
[0059] For example, in some embodiments, the overdue stage prediction model can predict the overdue stage of a user's next overdue period as follows: overdue for 2 months, overdue for 2 quarters, overdue for 1 year, etc. There is no limitation here, and it can be flexibly set according to the actual application scenario.
[0060] In this embodiment, the overdue stage prediction model adopts a bidirectional LSTM network. After inputting the overdue features of each historical overdue stage into the overdue stage prediction model, the overdue stage prediction model can comprehensively capture the user's overdue features from both positive and negative directions, thereby enhancing the prediction accuracy.
[0061] As can be seen, the overdue stage prediction method provided in this embodiment of the invention, by predicting users' overdue status based on the statistical characteristics of overdue stages, significantly reduces the amount of data processing and improves prediction efficiency compared to collecting users' daily data. Furthermore, the use of a bidirectional LSTM network allows for comprehensive capture of users' overdue characteristics from two directions, enhancing prediction accuracy. In addition, using stage-based features not only provides an understandable basis for explaining subsequent collection strategies but also simplifies the time cost of understanding the importance of collection features.
[0062] In one implementation, the bidirectional LSTM network used in the overdue stage prediction model includes a forward LSTM network and a reverse LSTM network. Specifically, step S102, inputting the overdue features from each historical overdue stage into the pre-trained overdue stage prediction model, may include:
[0063] The overdue features within each historical overdue stage are input into the forward LSTM network of the overdue stage prediction model in a first order, and the overdue features within each historical overdue stage are input into the inverse LSTM network of the overdue stage prediction model in a second order; the first order and the second order are determined according to the order of overdue payments corresponding to each historical overdue stage, and the first order and the second order are opposite.
[0064] In this embodiment, the first order can be from the earliest overdue record to the most recent overdue, and the second order can be the order of tracing back from the most recent overdue. For example, if a user has three overdue experiences in the past, and the overdue periods from the earliest overdue record to the most recent overdue are m2 (assuming 2 months), m1 (assuming 1 month), and m3 (assuming 3 months), then after obtaining the overdue features of the three overdue stages m2, m1, and m3, the corresponding overdue features will be input into the forward LSTM network in the order of m2, m1, and m3, and the corresponding overdue features will be input into the inverse LSTM network in the order of m3, m1, and m2.
[0065] As can be seen, the present invention employs a bidirectional LSTM network. By combining forward and reverse time series information, the model can learn more features, thereby making more accurate predictions and improving the prediction accuracy for users' future overdue periods.
[0066] In one implementation, the overdue stage prediction model can be trained in the following manner:
[0067] A training dataset is generated based on the overdue characteristics of different users across multiple historical overdue periods; the overdue period prediction model based on a bidirectional LSTM network is then trained using the training dataset to obtain the trained overdue period prediction model.
[0068] Optionally, the training dataset can be obtained from transaction data of a bank or financial institution within a preset historical period.
[0069] In some embodiments, the overdue characteristics may include: basic user information, historical credit, behavioral data, and financial data, etc., which are not limited herein.
[0070] Optionally, basic user information may include: age, occupation, income level, education level, etc.; historical credit may include: past loan records, repayment performance, number of overdue payments, etc.; behavioral data may include: APP login frequency, page dwell time, information modification behavior, etc.; financial data may include: debt-to-income ratio, account transaction history, asset certificates, etc. There are no restrictions here, and they can be flexibly set according to the actual application scenario.
[0071] Suppose a user has four overdue payments in their history. We need to statistically analyze the overdue characteristics of these four overdue periods and input them into the model for training. The overdue periods corresponding to the four overdue payments are m2, m1, m3, and m1. Taking the overdue characteristics of the first overdue period as an example, we statistically analyze the following features within the first overdue period: number of overdue days, overdue amount, outstanding amount, loan app usage clicks, car GPS speed, distance statistics, etc. We then generate training data and label each training data point (i.e., the user's actual overdue period for the next overdue payment), thus constructing the training dataset.
[0072] When training the model, we can take the training data of overdue features of the four overdue stages m2, m1, m3, and m1 as an example. The corresponding overdue features can be input into the forward LSTM network in the order of m2, m1, m3, and m1, and the corresponding overdue features can be input into the inverse LSTM network in the order of m1, m3, m1, and m2. The overdue stage prediction model is trained until the overdue stage prediction model achieves the expected effect.
[0073] As can be seen, in the embodiments of the present invention, the features are statistically analyzed over time during the overdue period when training the model, which reduces the data explosion caused by statistical analysis of features on a daily basis, shortens the time required for model training, and saves a lot of equipment resources.
[0074] In one implementation, to help business personnel better understand the model's decision-making process and ensure the model's high interpretability, this embodiment of the invention introduces the LRP algorithm to evaluate the importance of each overdue feature in the model's input layer. This clearly shows which overdue features have the greatest impact on the prediction results, thereby improving the model's interpretability and providing a more accurate interpretable basis for subsequent collection strategy optimization.
[0075] Referring to Figure 2, the overdue stage prediction method provided in this embodiment of the invention may further include:
[0076] Step S201: After the overdue stage prediction model is trained, the LRP algorithm is used to evaluate the importance of each overdue feature.
[0077] In this embodiment, the LRP algorithm is mainly used for model interpretation and feature importance assessment. That is, after the model has completed its prediction, it is used to understand the model's decision-making process, thereby determining which input features have the greatest impact on the final prediction result. For example, a user's historical overdue records include features such as overdue days, overdue amount, and app usage habits. When using a trained overdue stage prediction model to predict the user's next overdue stage, the LRP algorithm can determine which features (such as overdue days or app usage frequency) have the greatest impact on the prediction result.
[0078] Optionally, after using the LRP algorithm to assess the importance of each overdue feature, the process further includes:
[0079] Optimize collection strategies for users based on the importance of each overdue characteristic.
[0080] Understandably, the importance of each overdue characteristic can indicate its impact on the overdue stage. The more important each overdue characteristic is, the greater its impact on the overdue stage; otherwise, its impact on the overdue stage is smaller.
[0081] In some embodiments, after obtaining the importance of each overdue characteristic, the collection strategy for the user can be optimized accordingly. Specifically, the collection time (e.g., changing the collection time from the beginning of the month to the middle of the month), collection method, etc. can be optimized. There are no limitations here, and they can be flexibly set according to the actual application scenario.
[0082] In one implementation, step S201 above uses the LRP algorithm to assess the importance of each overdue feature, which may specifically include:
[0083] Based on the correlation of the current layer in the overdue stage prediction model, the correlation calculation results from the current layer to the previous layer of the forward LSTM network and the inverse LSTM network in the overdue stage prediction model are calculated respectively. The correlation calculation results from the current layer to the previous layer of the forward LSTM network and the inverse LSTM network are weighted to obtain the correlation of the previous layer of the current layer in the overdue stage prediction model. Through layer-by-layer iterative calculation, the importance of each overdue feature is finally output through the overdue stage prediction model.
[0084] In this embodiment, when using the LRP algorithm to assess the importance of overdue features, the calculation starts from the output layer of the overdue stage prediction model and iterates layer by layer, ultimately outputting the correlation of each overdue feature through the overdue stage prediction model. The correlation of the output layer of the overdue stage prediction model can be preset.
[0085] In this embodiment, the formula for calculating the correlation between the current layer and the previous layer in a forward LSTM network is as follows:
[0086] Where i represents the previous layer, j represents the current layer, and R i←j R represents the correlation calculation result between the current layer and the previous layer in a forward LSTM network. j z represents the relevance of the current layer of the overdue stage prediction model. i w represents the output of the i-th layer of a forward LSTM network. ijThis represents the weight of the connection between layers i and j in a forward LSTM network.
[0087] The formula for calculating the correlation between the current layer and the previous layer in a reverse LSTM network is as follows:
[0088] Among them, R′ i←j This represents the correlation calculation result between the current layer and the previous layer of the inverse LSTM network, z′. i w′ represents the output of the i-th layer of the inverse LSTM network. ij This represents the weight of the connection between the i-th and j-th layers of the reverse LSTM network.
[0089] In this embodiment, the correlation R between the current layer and the previous layer of the forward LSTM network in the overdue stage prediction model is calculated. i←j And the correlation calculation result R′ of the current layer to the previous layer of the inverse LSTM network in the overdue stage prediction model. i←j Then, through weighted calculation, the correlation R of the previous layer of the current layer in the overdue stage prediction model is finally obtained. i Specifically, it can be represented as R i =A*R i←j +B*R′ i←j A and B are preset weight values. For example, you can set A = 0.5 and B = 0.5.
[0090] The correlation R of the previous layer of the current layer in the overdue stage prediction model is obtained. i Then, through iterative calculation, the importance of each overdue feature in the input layer is obtained, thereby determining the importance of each overdue feature to the current prediction result.
[0091] As can be seen, the embodiments of the present invention propose a bidirectional LRP interpretation method for the overdue stage prediction model based on bidirectional LSTM network. This method can reduce the problem of forgetting the importance of earlier features when interpreting unidirectional time series, provide accurate interpretability basis for subsequent collection, and reduce the manual cost of updating the strategy.
[0092] To perform the corresponding steps in the above embodiments and various possible methods, an implementation of a delinquency stage prediction device is given below. Please refer to Figure 3, which is a functional block diagram of a delinquency stage prediction device 600 provided in an embodiment of the present invention. It should be noted that the basic principle and technical effects of the delinquency stage prediction device 600 provided in this embodiment are the same as those in the above embodiments. For the sake of brevity, any parts not mentioned in this embodiment can be referred to the corresponding content in the above embodiments. The delinquency stage prediction device 600 includes: an acquisition module 610, an input module 620, and a prediction module 630.
[0093] The acquisition module 610 is used to acquire the overdue characteristics of the user in each historical overdue period; the historical overdue period includes multiple consecutive days.
[0094] It is understood that the acquisition module 610 can perform the above step S101.
[0095] The input module 620 is used to input the overdue features of each historical overdue stage into the pre-trained overdue stage prediction model; the overdue stage prediction model adopts a bidirectional LSTM network.
[0096] It is understood that the input module 620 can perform the above step S102.
[0097] The prediction module 630 is used to predict the next overdue stage of a user through an overdue stage prediction model.
[0098] It is understood that the prediction module 630 can perform the above step S103.
[0099] Optionally, the input module 620 is used to input the overdue features in each historical overdue stage into the forward LSTM network of the overdue stage prediction model in a first order, and input the overdue features in each historical overdue stage into the inverse LSTM network of the overdue stage prediction model in a second order; the first order and the second order are determined according to the overdue order corresponding to each historical overdue stage, and the first order and the second order are opposite.
[0100] Optionally, the overdue stage prediction model is trained in the following manner:
[0101] A training dataset is generated based on the overdue characteristics of different users across multiple historical overdue periods.
[0102] The overdue stage prediction model based on the bidirectional LSTM network is trained using the training dataset to obtain the trained overdue stage prediction model.
[0103] Optionally, referring to Figure 4, the overdue stage prediction device 600 may further include a calculation module 640, which is used to evaluate the importance of each overdue feature using the LRP algorithm after the overdue stage prediction model has been trained.
[0104] It is understood that the calculation module 640 can perform the above step S201.
[0105] Optionally, the calculation module 640 is specifically used to calculate the correlation between the current layer and the previous layer of the forward LSTM network in the overdue stage prediction model, and the correlation between the current layer and the previous layer of the inverse LSTM network in the overdue stage prediction model, based on the correlation of the current layer of the overdue stage prediction model; to perform weighted calculation on the correlation between the current layer and the previous layer of the forward LSTM network and the correlation between the current layer and the previous layer of the inverse LSTM network to obtain the correlation of the previous layer of the current layer of the overdue stage prediction model; and to finally output the importance of each overdue feature through layer-by-layer iterative calculation.
[0106] Optionally, the formula for calculating the correlation between the current layer and the previous layer in a forward LSTM network is as follows:
[0107] Where i represents the previous layer, j represents the current layer, and R i←j R represents the correlation calculation result between the current layer and the previous layer in a forward LSTM network. j z represents the relevance of the current layer of the overdue stage prediction model. i w represents the output of the i-th layer of a forward LSTM network. ij This represents the weight of the connection between the i-th and j-th layers of a forward LSTM network.
[0108] The formula for calculating the correlation between the current layer and the previous layer in a reverse LSTM network is as follows:
[0109] Among them, R′ i←j This represents the correlation calculation result between the current layer and the previous layer of the inverse LSTM network, z′. i w′ represents the output of the i-th layer of the inverse LSTM network. ij This represents the weight of the connection between the i-th and j-th layers of the reverse LSTM network.
[0110] As can be seen, the overdue stage prediction device provided in this embodiment of the invention includes an acquisition module, an input module, and a prediction module. The acquisition module is used to acquire the overdue characteristics of the user in each historical overdue stage; the historical overdue stage includes multiple consecutive days. The input module is used to input the overdue characteristics of each historical overdue stage into a pre-trained overdue stage prediction model; the overdue stage prediction model adopts a bidirectional LSTM network. The prediction module is used to predict the overdue stage in which the user will be overdue next through the overdue stage prediction model. By predicting the user's overdue situation based on the statistical characteristics of overdue stages, compared with collecting the user's daily data, the amount of data processing is greatly reduced and the prediction efficiency is improved. Furthermore, by adopting a bidirectional LSTM network, the user's overdue characteristics can be captured comprehensively from two directions, enhancing the prediction accuracy.
[0111] Please refer to Figure 5, which is a block diagram of an electronic device provided in an embodiment of the present invention. The electronic device 100 includes a memory 110, a processor 120, and a communication module 130. The memory 110, processor 120, and communication module 130 are electrically connected to each other directly or indirectly to achieve data transmission or interaction. For example, these components can be electrically connected to each other through one or more communication buses or signal lines.
[0112] The memory 110 is used to store programs or data. The memory 110 may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.
[0113] The processor 120 is used to read / write data or programs stored in the memory 110 and perform corresponding functions. For example, when the computer program stored in the memory 110 is executed by the processor 120, the overdue stage prediction method disclosed in the above embodiments can be implemented.
[0114] The communication module 130 is used to establish a communication connection between the electronic device 100 and other devices via a network, and to send and receive data via the network.
[0115] It should be understood that the structure shown in Figure 5 is only a schematic diagram of the electronic device 100. The electronic device 100 may also include more or fewer components than those shown in Figure 5, or have a different configuration than those shown in Figure 5. The components shown in Figure 5 may be implemented using hardware, software, or a combination thereof.
[0116] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor 120, implements the overdue stage prediction method disclosed in the above embodiments.
[0117] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative; for example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.
[0118] In addition, the functional modules in the various embodiments of the present invention can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.
[0119] If the functionality is implemented as a software module 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 invention, or the part that contributes to the prior art, or a 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 of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0120] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention. Industrial applicability
[0121] By adopting the above scheme, the amount of data processing is greatly reduced and the prediction efficiency is improved compared to collecting daily data from users, by using statistical characteristics of overdue stages. Furthermore, the use of a bidirectional LSTM network can comprehensively capture the overdue characteristics of users from two directions, thereby enhancing the prediction accuracy.
Claims
1. A method for predicting overdue periods, characterized in that, The method includes: Obtain the overdue characteristics of the user in each historical overdue period; the historical overdue period includes multiple consecutive days; The overdue features within each of the aforementioned historical overdue stages are input into a pre-trained overdue stage prediction model; the overdue stage prediction model employs a bidirectional LSTM network. The overdue stage prediction model predicts the overdue stage at which a user will be overdue next.
2. The overdue stage prediction method according to claim 1, characterized in that, The step of inputting the overdue features within each of the aforementioned historical overdue stages into a pre-trained overdue stage prediction model includes: The overdue features within each of the historical overdue stages are input into the forward LSTM network of the overdue stage prediction model in a first order, and the overdue features within each of the historical overdue stages are input into the inverse LSTM network of the overdue stage prediction model in a second order; the first order and the second order are determined according to the order of overdue periods corresponding to each of the historical overdue stages, and the first order and the second order are reversed.
3. The overdue stage prediction method according to claim 1 or 2, characterized in that, The overdue stage prediction model was trained in the following manner: A training dataset is generated based on the overdue characteristics of different users in multiple historical overdue stages, and the training dataset is labeled with the actual overdue stage of the next overdue payment. The overdue stage prediction model based on the bidirectional LSTM network is trained using the training dataset to obtain the trained overdue stage prediction model.
4. The overdue stage prediction method according to any one of claims 1-3, characterized in that, The method further includes: After the overdue stage prediction model is trained, the LRP algorithm is used to evaluate the importance of each overdue feature.
5. The overdue stage prediction method according to claim 4, characterized in that, The use of the LRP algorithm to assess the importance of each overdue feature includes: Based on the correlation of the current layer of the overdue stage prediction model, the correlation calculation results of the current layer to the previous layer of the forward LSTM network in the overdue stage prediction model and the correlation calculation results of the current layer to the previous layer of the reverse LSTM network in the overdue stage prediction model are calculated respectively. The importance of each overdue feature is calculated based on the correlation calculation results of the current layer to the previous layer of the forward LSTM network in the overdue stage prediction model and the correlation calculation results of the current layer to the previous layer of the reverse LSTM network in the overdue stage prediction model.
6. The overdue stage prediction method according to claim 5, characterized in that, The importance of each overdue feature is calculated based on the correlation calculation results of the current layer to the previous layer of the forward LSTM network in the overdue stage prediction model and the correlation calculation results of the current layer to the previous layer of the inverse LSTM network in the overdue stage prediction model, including: The correlation calculation results from the current layer to the previous layer of the forward LSTM network and the correlation calculation results from the current layer to the previous layer of the reverse LSTM network are weighted and calculated to obtain the correlation of the current layer to the previous layer of the overdue stage prediction model. Through iterative calculations layer by layer, the correlation of each overdue feature is finally output through the overdue stage prediction model.
7. The overdue stage prediction method according to claim 5 or 6, characterized in that, The formula for calculating the correlation between the current layer and the previous layer in the forward LSTM network is as follows: Where i represents the previous layer, j represents the current layer, and R i←j R represents the correlation calculation result from the current layer to the previous layer of the forward LSTM network. j The z represents the correlation of the current layer of the overdue stage prediction model. i w represents the output of the i-th layer of the forward LSTM network. ij This represents the weight of the connection between the i-th and j-th layers of the forward LSTM network.
8. The overdue stage prediction method according to any one of claims 5-7, characterized in that, The formula for calculating the correlation between the current layer and the previous layer of the reverse LSTM network is as follows: Among them, R′ i←j This represents the correlation calculation result from the current layer to the previous layer of the inverse LSTM network, z′. i w′ represents the output of the i-th layer of the inverse LSTM network. ij This represents the weight of the connection between the i-th and j-th layers of the reverse LSTM network.
9. The overdue stage prediction method according to any one of claims 4-8, characterized in that, After using the LRP algorithm to assess the importance of each overdue feature, the process also includes: Optimize collection strategies for the users based on the importance of each overdue characteristic.
10. The method for predicting overdue stages according to any one of claims 1-9, characterized in that, The overdue phase is used to indicate the duration of the overdue period.
11. A device for predicting overdue periods, characterized in that, The device includes: The acquisition module is used to acquire the overdue characteristics of a user in each historical overdue period; the historical overdue period includes multiple consecutive days; The input module is used to input the overdue features of each of the historical overdue stages into a pre-trained overdue stage prediction model; the overdue stage prediction model adopts a bidirectional LSTM network. The prediction module is used to predict the next overdue stage of a user through the overdue stage prediction model.
12. The overdue stage prediction device according to claim 11, characterized in that, The input module is used to input the overdue features in each of the historical overdue stages into the forward LSTM network of the overdue stage prediction model in a first order, and to input the overdue features in each of the historical overdue stages into the inverse LSTM network of the overdue stage prediction model in a second order; the first order and the second order are determined according to the overdue order corresponding to each of the historical overdue stages, and the first order and the second order are opposite.
13. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the overdue stage prediction method as described in any one of claims 1-10.
14. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, which, when executed by a processor, implements the steps of the overdue stage prediction method as described in any one of claims 1-10.