A risk prediction window selection method and device based on multi-target collaborative evaluation and electronic equipment

Abstract

The application provides a risk prediction window selection method and device based on multi-target cooperative evaluation and electronic equipment, relates to the technical field of machine learning and data modeling, and comprises the following steps: obtaining time sequence historical data of a target object, and dividing the time sequence historical data into a training subset and a test subset used for simulating future prediction according to time sequence; a candidate prediction window set containing at least two different time lengths is set, for each window length in the set, a corresponding binary classification prediction task label is defined based on the length, and a corresponding prediction model is trained by using the training subset; taking the model performance corresponding to the longest reference window in the candidate window as a reference, the optimal window length is selected from the candidate window set according to a preset selection criterion. The application ensures that the selected window length directly serves the accuracy target of the final prediction model, so that the shortest effective window is automatically found under the premise of ensuring that the prediction performance of the long window reference model is equivalent.

Description

Technical Field

[0001] This invention relates to the fields of machine learning and data modeling technology, and in particular to a risk prediction window selection method, apparatus and electronic device based on multi-objective collaborative assessment. Background Technology

[0002] In risk assessment or behavioral prediction modeling based on historical data to predict specific future states, selecting the prediction window length is a crucial preliminary step. Existing technologies mainly fall into two categories: Rule-based static window setting: This method sets a fixed window length based on domain experience or general specifications. While this ensures the standardization of the process, it lacks adaptability to specific business scenarios and data characteristics, which can easily lead to poor model timeliness and an inability to capture recent state change patterns.

[0003] The dynamic window derivation method based on statistical regularities derives the window length by analyzing the statistical regularities of target events in historical data (such as state transition probabilities and survival analysis). While this method is data-driven, its derivation process relies on multiple statistical assumptions, is susceptible to data noise, and may lead to a disconnect between the derivation results and the actual predictive performance of the model, posing a risk of overfitting.

[0004] A common drawback of existing technologies is that the choice of window length is independent of the final prediction model's performance optimization process. They rely on indirect prior rules or intermediate statistics, rather than directly optimizing to improve the model's core performance metrics (such as discrimination and calibration) on the prediction target. This leads to the selected window not being the optimal solution for model performance, often causing the following problems: high data acquisition and storage costs, long model update cycles, insufficient capture of recent dynamic patterns, and decreased model generalization ability due to mismatched window lengths.

[0005] Therefore, a risk prediction window selection method, device, and electronic device based on multi-objective collaborative assessment are proposed. Summary of the Invention

[0006] This specification provides a risk prediction window selection method, apparatus, and electronic device based on multi-objective collaborative assessment, which ensures that the selected window length directly serves the accuracy target of the final prediction model, thereby automatically finding the shortest effective window while ensuring prediction performance comparable to that of a long-window benchmark model.

[0007] This specification provides a risk prediction window selection method based on multi-objective collaborative assessment, including: Obtain the time-series historical data of the target object and divide it into a training subset and a test subset for simulating future predictions in chronological order; Set a set of candidate prediction windows containing at least two different time lengths. For each window length in the set, define a corresponding binary classification prediction task label based on the length, and train a corresponding prediction model using the training subset. All models use the same feature input. On the test subset, the performance of each of the prediction models is evaluated, and the performance evaluation includes at least two dimensions of quantitative metrics. Based on the performance of the model corresponding to the longest baseline window in the candidate window set, the optimal window length is selected from the candidate window set according to a preset selection criterion. The selection criterion is: to find the first window length from shortest to longest whose corresponding model has a performance difference of less than the baseline model in multiple performance indicators.

[0008] Optionally, the chronological division into a training subset and a test subset for simulating future predictions includes: Data is divided into training subsets based on a specific time point as the cut-off point, with the data before the cut-off point constituting the training subset and the data after the cut-off point constituting the test subset.

[0009] Optionally, the prediction model employs a machine learning model based on a gradient boosting decision tree framework, and is optimized using a gradient descent algorithm with the objective of minimizing the binary classification cross-entropy loss. Specifically: in, The loss function; The number of samples; This is the actual label, and its value is either 0 or 1; This represents the probability predicted by the model.

[0010] Optionally, the setting includes at least two sets of candidate prediction windows with different time lengths. For each window length in the set, a corresponding binary classification prediction task label is defined based on that length, and a corresponding prediction model is trained using the training subset. All models use the same feature input, including: When training the prediction model using a training subset, a Bayesian optimization method is used to optimize the combination of hyperparameters of the model. The Bayesian optimization constructs a surrogate model of the objective function and performs a guided search in the hyperparameter space using the acquisition function to obtain the optimal combination of hyperparameters for the validation set within a finite number of evaluations. Specifically: in, The optimal combination of hyperparameters; For hyperparameter combinations; For hyperparameter space; This is the data acquisition function; This is historical observation data.

[0011] Optionally, on the test subset, the performance of each prediction model is evaluated. This performance evaluation includes at least two quantitative metrics: the area under the receiver operating characteristic curve (AUC) and / or the Kolmokolov-Smirnov statistic, which assess the model's discriminative ability; and the cross-entropy loss value, which assesses the accuracy of the predicted probability calibration. Specifically: in, This indicates the performance period. The area under the operating feature curve of the trained model on the test set; For the true rate; The false positive rate; in, This indicates the performance period. The Kolmokolov-Smirnov statistic of the trained model on the test set; Let be the cumulative distribution function of the predicted scores for positive samples; The cumulative distribution function of the predicted scores for negative samples; in, This represents the cross-entropy loss value. For the performance period The trained model on the samples The predicted probability.

[0012] Optionally, the step of using the model performance corresponding to the longest baseline window among the candidate windows as the benchmark includes: The performance difference function between each indicator and the final performance period, specifically: in, The AUC difference function; The KS difference function; This is the cross-entropy difference function.

[0013] Optional, also includes: After determining the theoretically optimal window length according to the selection criteria, the actual modeling process uses a window length that is 1 to 2 time periods shorter than the theoretically optimal window length for model training and updating.

[0014] This specification provides a risk prediction window selection device based on multi-objective collaborative assessment, including: The acquisition module is used to acquire the time-series historical data of the target object and divide it into a training subset and a test subset for simulating future predictions in chronological order. The training module is used to set a set of candidate prediction windows containing at least two different time lengths. For each window length in the set, a corresponding binary classification prediction task label is defined based on the length, and a corresponding prediction model is trained using the training subset. All models use the same feature input. An evaluation module is used to evaluate the performance of each of the prediction models on the test subset, wherein the performance evaluation includes at least two-dimensional quantitative metrics. The optimization module is used to select the optimal window length from the candidate window set based on the model performance corresponding to the longest benchmark window in the candidate window and according to a preset selection criterion. The selection criterion is to find the first window length from shortest to longest whose corresponding model has a performance difference from the benchmark model in multiple dimensions that is less than their respective preset thresholds.

[0015] Optionally, the acquisition module includes: Data is divided into training subsets based on a specific time point as the cut-off point, with the data before the cut-off point constituting the training subset and the data after the cut-off point constituting the test subset.

[0016] Optionally, the prediction model employs a machine learning model based on a gradient boosting decision tree framework, and is optimized using a gradient descent algorithm with the objective of minimizing the binary classification cross-entropy loss. Specifically: in, The loss function; The number of samples; This is the actual label, and its value is either 0 or 1; This represents the probability predicted by the model.

[0017] Optionally, the training module includes: When training the prediction model using a training subset, a Bayesian optimization method is used to optimize the combination of hyperparameters of the model. The Bayesian optimization constructs a surrogate model of the objective function and performs a guided search in the hyperparameter space using the acquisition function to obtain the optimal combination of hyperparameters for the validation set within a finite number of evaluations. Specifically: in, The optimal combination of hyperparameters; For hyperparameter combinations; For hyperparameter space; This is the data acquisition function; This is historical observation data.

[0018] Optionally, the evaluation module includes: the area under the receiver operating characteristic curve and / or the Kolmokolov-Smirnov statistic to assess the model's discriminative ability, and the cross-entropy loss value to assess the accuracy of the predicted probability calibration, specifically: in, This indicates the performance period. The area under the operating feature curve of the trained model on the test set; For the true rate; The false positive rate; in, This indicates the performance period. The Kolmokolov-Smirnov statistic of the trained model on the test set; Let be the cumulative distribution function of the predicted scores for positive samples; The cumulative distribution function of the predicted scores for negative samples; in, This represents the cross-entropy loss value. For the performance period The trained model on the samples The predicted probability.

[0019] Optionally, the selection module includes: The performance difference function between each indicator and the final performance period, specifically: in, The AUC difference function; The KS difference function; This is the cross-entropy difference function.

[0020] Optional, also includes: After determining the theoretically optimal window length according to the selection criteria, the actual modeling process uses a window length that is 1 to 2 time periods shorter than the theoretically optimal window length for model training and updating.

[0021] This specification also provides an electronic device, wherein the electronic device includes: A processor; and a memory storing computer-executable instructions, which, when executed, cause the processor to perform any of the methods described above.

[0022] This specification also provides a computer-readable storage medium that stores one or more programs that, when executed by a processor, implement any of the methods described above.

[0023] In this invention, by independently modeling different candidate windows and directly performing multi-dimensional performance evaluation and comparison on a unified future test set, it is ensured that the selected window length directly serves the accuracy target of the final prediction model. This automatically finds the shortest effective window while maintaining predictive performance comparable to long-window benchmark models. This fundamentally solves the problems of high data costs, slow model updates, and significant time lags caused by long windows, as well as the potential for insufficient prediction accuracy from short windows or simple rules. Rigorous out-of-time validation, unified feature input, and a high-standard model training process based on gradient boosting decision trees and Bayesian optimization provide a solid and reliable empirical foundation for window selection. The use of a clear, quantified performance difference function and a pre-set threshold-based multi-objective collaborative selection criterion makes the decision-making process objective, transparent, and reproducible, effectively avoiding subjective experience interference and overfitting risks caused by complex statistical assumptions, significantly improving the robustness and generalization ability of the solution. The optimal window determined in this invention is typically a significantly shorter, more efficient window than traditional long windows (such as 12 periods). This directly reduces the length and time span of historical data required for modeling, thereby saving data storage and computing resources and significantly accelerating the iteration speed of the model from data preparation to training and deployment. This allows the model to respond more agilely to the latest changes in market and user behavior. Furthermore, utilizing more recent data often yields dynamic adaptability superior to static theory in practice, enhancing the model's practical value in production environments. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 A schematic diagram illustrating the principle of a risk prediction window selection method based on multi-objective collaborative assessment provided in the embodiments of this specification; Figure 2 A schematic diagram of a risk prediction window selection device based on multi-objective collaborative assessment provided in the embodiments of this specification; Figure 3This is a schematic diagram of the structure of an electronic device provided in an embodiment of this specification; Figure 4 This is a schematic diagram of a computer-readable medium provided for embodiments of this specification. Detailed Implementation

[0026] The following description is intended to disclose the present invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art. The basic principles of the invention defined in the following description can be applied to other embodiments, modifications, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the invention.

[0027] The following is in conjunction with the appendix Figure 1-4 Exemplary embodiments of the invention will be described more fully here. However, exemplary embodiments can be implemented in many forms and should not be construed as limiting the invention to the embodiments set forth herein. Rather, these exemplary embodiments are provided to make the invention more comprehensive and complete, and to facilitate a full communication of the inventive concept to those skilled in the art. The same reference numerals in the figures denote the same or similar elements, components, or parts, and therefore repeated descriptions of them are omitted.

[0028] Subject to the technical concept of this invention, the features, structures, characteristics or other details described in a particular embodiment may be combined in one or more other embodiments in a suitable manner.

[0029] In the description of specific embodiments, the features, structures, characteristics, or other details described in this invention are intended to enable those skilled in the art to fully understand the embodiments. However, it is not excluded that those skilled in the art can practice the technical solutions of this invention without one or more of the specific features, structures, characteristics, or other details.

[0030] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0031] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0032] The terms “and / or” or “and / or” include all combinations of any one or more of the listed items.

[0033] Figure 1 This specification provides a schematic diagram illustrating the principle of a risk prediction window selection method based on multi-objective collaborative assessment, which may include: S110: Obtain the time-series historical data of the target object and divide it into a training subset and a test subset for simulating future predictions in chronological order; Optionally, S110 includes: Data is divided into training subsets based on a specific time point as the cut-off point, with the data before the cut-off point constituting the training subset and the data after the cut-off point constituting the test subset.

[0034] In the specific implementation of this specification, firstly, the time-series historical data of the target object (such as individual users, enterprise entities, etc.) is acquired. This data is strictly divided in chronological order, with a certain point in time as the cut-off date. The data before the cut-off date is divided into a training subset for model building; the data after the cut-off date is divided into a test subset (i.e., the out-of-time test set), which is specifically used to simulate the future and evaluate the generalization performance of the model.

[0035] S120: Set a set of candidate prediction windows containing at least two different time lengths. For each window length in the set, define a corresponding binary classification prediction task label based on the length, and train a corresponding prediction model using the training subset. All models use the same feature input. Optionally, the prediction model employs a machine learning model based on a gradient boosting decision tree framework, and is optimized using a gradient descent algorithm with the objective of minimizing the binary classification cross-entropy loss. Specifically: in, The loss function; The number of samples; This is the actual label, and its value is either 0 or 1; This represents the probability predicted by the model.

[0036] In the specific implementation of this specification, the prediction model selected is a machine learning model based on the Gradient Boosting Decision Tree (GBDT) framework, such as LightGBM. This model is iteratively optimized using the gradient descent algorithm, with the training objective being to minimize the binary cross-entropy loss. The loss function effectively measures the difference between the model's predicted probability distribution and the true label distribution, guiding the model to optimize towards more accurate predictions and more calibrated probabilities.

[0037] Optionally, S120 includes: When training the prediction model using a training subset, a Bayesian optimization method is used to optimize the combination of hyperparameters of the model. The Bayesian optimization constructs a surrogate model of the objective function and performs a guided search in the hyperparameter space using the acquisition function to obtain the optimal combination of hyperparameters for the validation set within a finite number of evaluations. Specifically: in, The optimal combination of hyperparameters; For hyperparameter combinations; For hyperparameter space; This is the data acquisition function; This is historical observation data.

[0038] In the specific implementation of this specification, to further improve the performance of each window model and achieve automation and efficiency in hyperparameter tuning, a Bayesian optimization method is introduced to optimize the hyperparameter combination of the model during training. Specifically, the core idea of ​​Bayesian optimization is to transform the hyperparameter optimization problem into a global optimization problem of a black-box function. It approximates the real, computationally expensive model evaluation process by sequentially constructing a surrogate model (e.g., a Gaussian process regression model) of the objective function (e.g., the model's AUC value on the validation set). Then, an acquisition function (e.g., expected improvement in EI, upper confidence bound UCB, etc.) is used to balance the relationship between "utilization" (searching in regions where the surrogate model performs well) and "exploration" (searching in regions with high uncertainty), thereby intelligently recommending the next hyperparameter combination to be evaluated. By iteratively executing this process (e.g., performing 50 rounds of evaluation), the hyperparameter combination that achieves optimal or near-optimal performance on the validation set can be efficiently found within a limited number of evaluations, thus significantly improving the final predictive performance of each window model.

[0039] S130: On the test subset, evaluate the performance of each of the prediction models, the performance evaluation including at least two-dimensional quantitative metrics; Optionally, S130 specifically includes: in, This indicates the performance period. The area under the operating feature curve of the trained model on the test set; For the true rate; The false positive rate; in, This indicates the performance period. The Kolmokolov-Smirnov statistic of the trained model on the test set; Let be the cumulative distribution function of the predicted scores for positive samples; The cumulative distribution function of the predicted scores for negative samples; in, This represents the cross-entropy loss value. For the performance period The trained model on the samples The predicted probability.

[0040] In the specific implementation of this specification, when evaluating model performance on the test subset, the present invention uses multi-dimensional quantitative indicators for comprehensive evaluation, specifically including the following two categories: the first category is indicators used to evaluate the model's discrimination ability, such as the area under the receiver operating characteristic curve and / or the Kolmokolov-Smirnov statistic; the second category is indicators used to evaluate the model's predictive probability calibration, namely the cross-entropy loss value.

[0041] S140: Based on the performance of the model corresponding to the longest baseline window in the candidate window, select the optimal window length from the candidate window set according to the preset selection criteria. The selection criteria are: find the first window length from shortest to longest whose corresponding model has a performance difference of less than the baseline model in multiple performance indicators.

[0042] Optionally, S140 includes: The performance difference function between each indicator and the final performance period, specifically: in, is the AUC difference function; is the KS difference function; is the cross-entropy difference function.

[0043] In the specific implementation of this specification, when selecting the implementation window, it is necessary to quantify the performance difference between each candidate window model and the benchmark model. Let the benchmark window length be K (usually the longest window in the candidate set, such as 12 periods), and its corresponding model be M K . For any candidate window length k (k < K), its corresponding model is M k . Define the following three performance difference functions to calculate the relative degradation degrees of the AUC, KS, and cross-entropy loss metrics respectively.

[0044] Optionally, it further includes: After determining the theoretical optimal window length according to the selection criterion, when actually modeling, a window length that is 1 to 2 time periods shorter than the theoretical optimal window length is used for model training and updating.

[0045] In the specific implementation of this specification, in actual business deployment, to balance the stability of model performance and the ability to quickly respond to recent data patterns, the theoretical optimal window length can be optimized in engineering. Specifically, assume that the theoretical optimal window length selected by the above criterion is k opt (such as 6 months). In subsequent model iterative training and updating, a window length that is 1 to 2 time periods shorter than k opt can be used (such as using 4 or 5 months). The advantages of this approach are as follows: 1) The time span of the training data used is closer, enabling faster capture of the latest changes in the market and user behavior; 2) It shortens the waiting time from data generation to being available for modeling, accelerating the model iteration cycle; 3) In most cases, due to the higher information density of recent data, this strategy often achieves better prediction effects in actual production than using the theoretical k opt , realizing an engineering leap from "static optimal" to "dynamic efficient".

[0046] In this invention, by independently modeling different candidate windows and directly performing multi-dimensional performance evaluation and comparison on a unified future test set, it is ensured that the selected window length directly serves the accuracy target of the final prediction model. This automatically finds the shortest effective window while maintaining predictive performance comparable to long-window benchmark models. This fundamentally solves the problems of high data costs, slow model updates, and significant time lags caused by long windows, as well as the potential for insufficient prediction accuracy from short windows or simple rules. Rigorous out-of-time validation, unified feature input, and a high-standard model training process based on gradient boosting decision trees and Bayesian optimization provide a solid and reliable empirical foundation for window selection. The use of a clear, quantified performance difference function and a pre-set threshold-based multi-objective collaborative selection criterion makes the decision-making process objective, transparent, and reproducible, effectively avoiding subjective experience interference and overfitting risks caused by complex statistical assumptions, significantly improving the robustness and generalization ability of the solution. The optimal window determined in this invention is typically a significantly shorter, more efficient window than traditional long windows (such as 12 periods). This directly reduces the length and time span of historical data required for modeling, thereby saving data storage and computing resources and significantly accelerating the iteration speed of the model from data preparation to training and deployment. This allows the model to respond more agilely to the latest changes in market and user behavior. Furthermore, utilizing more recent data often yields dynamic adaptability superior to static theory in practice, enhancing the model's practical value in production environments.

[0047] Figure 2 A schematic diagram of a risk prediction window selection device based on multi-objective collaborative assessment provided in this specification embodiment, the device may include: The acquisition module 10 is used to acquire the time-series historical data of the target object and divide it into a training subset and a test subset for simulating future predictions in chronological order. Training module 20 is used to set a set of candidate prediction windows containing at least two different time lengths. For each window length in the set, a corresponding binary classification prediction task label is defined based on the length, and a corresponding prediction model is trained using the training subset. All models use the same feature input. Evaluation module 30 is used to evaluate the performance of each of the prediction models on the test subset, the performance evaluation including at least two-dimensional quantitative metrics; The selection module 40 is used to select the optimal window length from the candidate window set based on the model performance corresponding to the longest benchmark window in the candidate window and according to a preset selection criterion. The selection criterion is to find the first window length from shortest to longest whose corresponding model has a performance difference from the benchmark model in multiple dimensions that is less than their respective preset thresholds.

[0048] Optionally, the acquisition module 10 includes: Data is divided into training subsets based on a specific time point as the cut-off point, with the data before the cut-off point constituting the training subset and the data after the cut-off point constituting the test subset.

[0049] Optionally, the prediction model employs a machine learning model based on a gradient boosting decision tree framework, and is optimized using a gradient descent algorithm with the objective of minimizing the binary classification cross-entropy loss. Specifically: in, The loss function; The number of samples; This is the actual label, and its value is either 0 or 1; This represents the probability predicted by the model.

[0050] Optionally, the training module 20 includes: When training the prediction model using a training subset, a Bayesian optimization method is used to optimize the combination of hyperparameters of the model. The Bayesian optimization constructs a surrogate model of the objective function and performs a guided search in the hyperparameter space using the acquisition function to obtain the optimal combination of hyperparameters for the validation set within a finite number of evaluations. Specifically: in, The optimal combination of hyperparameters; For hyperparameter combinations; For hyperparameter space; This is the data acquisition function; This is historical observation data.

[0051] Optionally, the evaluation module 30 includes: the area under the receiver operating characteristic curve and / or the Kolmokolov-Smirnov statistic to assess the model's discriminative ability, and the cross-entropy loss value to assess the accuracy of the predicted probability calibration, specifically: in, This indicates the performance period. The area under the operating feature curve of the trained model on the test set; For the true rate; The false positive rate; in, This indicates the performance period. The Kolmokolov-Smirnov statistic of the trained model on the test set; Let be the cumulative distribution function of the predicted scores for positive samples; The cumulative distribution function of the predicted scores for negative samples; in, This represents the cross-entropy loss value. For the performance period The trained model on the samples The predicted probability.

[0052] Optionally, the selection module 40 includes: The performance difference function between each indicator and the final performance period, specifically: in, The AUC difference function; The KS difference function; This is the cross-entropy difference function.

[0053] Optional, also includes: After determining the theoretically optimal window length according to the selection criteria, the actual modeling process uses a window length that is 1 to 2 time periods shorter than the theoretically optimal window length for model training and updating.

[0054] The functions of the apparatus in this embodiment have been described in the above method embodiments. Therefore, for any parts not detailed in this embodiment, please refer to the relevant descriptions in the foregoing embodiments, which will not be repeated here.

[0055] Based on the same inventive concept, embodiments of this specification also provide an electronic device.

[0056] The following describes embodiments of the electronic device of the present invention, which can be considered as specific implementations of the methods and apparatus embodiments of the present invention described above. Details described in the embodiments of the electronic device of the present invention should be considered as supplements to the methods or apparatus embodiments described above; details not disclosed in the embodiments of the electronic device of the present invention can be implemented with reference to the methods or apparatus embodiments described above.

[0057] Figure 3 This is a schematic diagram of an electronic device provided as an embodiment of this specification. Refer to the following... Figure 3 The electronic device 300 according to this embodiment of the present invention will be described. Figure 3 The electronic device 300 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0058] like Figure 3As shown, the electronic device 300 is presented in the form of a general-purpose computing device. The components of the electronic device 300 may include, but are not limited to: at least one processing unit 310, at least one storage unit 320, a bus 330 connecting different system components (including storage unit 320 and processing unit 310), a display unit 340, etc.

[0059] The storage unit stores program code that can be executed by the processing unit 310, causing the processing unit 310 to perform the steps described in the processing method section of this specification according to various exemplary embodiments of the present invention. For example, the processing unit 310 can perform, for example... Figure 1 The steps are shown.

[0060] The storage unit 320 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 3201 and / or a cache storage unit 3202, and may further include a read-only memory unit (ROM) 3203.

[0061] The storage unit 320 may also include a program / utility 3204 having a set (at least one) program module 3205, such program module 3205 including but not limited to: an operating system, one or more application programs, other program modules and program data, each or some combination of these examples may include an implementation of a network environment.

[0062] Bus 330 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the various bus structures.

[0063] Electronic device 300 can also communicate with one or more external devices 400 (e.g., keyboard, pointing device, Bluetooth device, etc.), and with one or more devices that enable viewers to interact with electronic device 300, and / or with any device that enables electronic device 300 to communicate with one or more other computing devices (e.g., router, modem, etc.). This communication can be performed via input / output (I / O) interface 350. Furthermore, electronic device 300 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 360. Network adapter 360 can communicate with other modules of electronic device 300 via bus 330. It should be understood that, although... Figure 3As not shown, other hardware and / or software modules may be used in conjunction with electronic device 300, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0064] Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described in this invention can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this invention can be embodied in the form of a software product, which can be stored in a computer-readable storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, or network device, etc.) to execute the method described above according to this invention. When the computer program is executed by a data processing device, it enables the computer-readable medium to implement the method described above, i.e.: as... Figure 1 The method shown.

[0065] Figure 4 This is a schematic diagram of a computer-readable medium provided for embodiments of this specification.

[0066] accomplish Figure 1 The computer program of the method shown can be stored on one or more computer-readable media. A computer-readable medium can be a readable signal medium or a readable storage medium. A readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0067] The computer-readable storage medium may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The readable storage medium may also be any readable medium other than a readable storage medium, capable of transmitting, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0068] Program code for performing the operations of this invention can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the audience's computing device, partially on the audience's device, as a standalone software package, partially on the audience's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the audience's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0069] In summary, the present invention can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that in practice, general-purpose data processing devices such as microprocessors or digital signal processors (DSPs) can be used to implement some or all of the functions of some or all of the components according to the embodiments of the present invention. The present invention can also be implemented as a device or apparatus program (e.g., a computer program and computer program product) for performing part or all of the methods described herein. Such programs implementing the present invention can be stored on a computer-readable medium or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

[0070] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the present invention is not inherently related to any specific computer, virtual device, or electronic device, and various general-purpose devices can also implement the present invention. The above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0071] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0072] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A risk prediction window selection method based on multi-objective collaborative assessment, characterized in that, include: Obtain the time-series historical data of the target object and divide it into a training subset and a test subset for simulating future predictions in chronological order; Set a set of candidate prediction windows containing at least two different time lengths. For each window length in the set, define a corresponding binary classification prediction task label based on the length, and train a corresponding prediction model using the training subset. All models use the same feature input. On the test subset, the performance of each of the prediction models is evaluated, and the performance evaluation includes at least two dimensions of quantitative metrics. Based on the model performance corresponding to the longest baseline window in the candidate window, the optimal window length is selected from the candidate window set according to the preset selection criteria. The selection criterion is: to find the first window length from shortest to longest in which the performance difference between the corresponding model and the benchmark model in multidimensional performance indicators is less than their respective preset thresholds.

2. The risk prediction window selection method based on multi-objective collaborative assessment as described in claim 1, characterized in that, The chronological division into a training subset and a test subset for simulating future predictions includes: Data is divided into training subsets based on a specific time point as the cut-off point, with the data before the cut-off point constituting the training subset and the data after the cut-off point constituting the test subset.

3. The risk prediction window selection method based on multi-objective collaborative assessment as described in claim 1, characterized in that, The prediction model employs a machine learning model based on a gradient boosting decision tree framework, and is optimized using a gradient descent algorithm with the objective of minimizing the binary classification cross-entropy loss. Specifically: in, The loss function; The number of samples; This is the actual label, and its value is either 0 or 1; This represents the probability predicted by the model.

4. The risk prediction window selection method based on multi-objective collaborative assessment as described in claim 3, characterized in that, The setting includes at least two sets of candidate prediction windows with different time lengths. For each window length in the set, a corresponding binary classification prediction task label is defined based on that length, and a corresponding prediction model is trained using the training subset. All models use the same feature input, including: When training the prediction model using a training subset, a Bayesian optimization method is used to optimize the combination of hyperparameters of the model. The Bayesian optimization constructs a surrogate model of the objective function and performs a guided search in the hyperparameter space using the acquisition function to obtain the optimal combination of hyperparameters for the validation set within a finite number of evaluations. Specifically: in, The optimal combination of hyperparameters; For hyperparameter combinations; For hyperparameter space; This is the data acquisition function; This is historical observation data.

5. The risk prediction window selection method based on multi-objective collaborative assessment as described in claim 1, characterized in that, On the test subset, the performance of each prediction model is evaluated. This performance evaluation includes at least two quantitative metrics: the area under the receiver operating characteristic curve (AUC) and / or the Kolmokolov-Smirnov statistic, which assess the model's discriminative ability; and the cross-entropy loss value, which assesses the accuracy of the predicted probability calibration. Specifically: in, This indicates the performance period. The area under the operating feature curve of the trained model on the test set; For the true rate; The false positive rate; in, This indicates the performance period. The Kolmokolov-Smirnov statistic of the trained model on the test set; Let be the cumulative distribution function of the predicted scores for positive samples; The cumulative distribution function of the predicted scores for negative samples; in, This represents the cross-entropy loss value. For the performance period The trained model on the samples The predicted probability.

6. The risk prediction window selection method based on multi-objective collaborative assessment as described in claim 5, characterized in that, The benchmarking of model performance based on the longest baseline window among candidate windows includes: The performance difference function between each indicator and the final performance period, specifically: in, The AUC difference function; The KS difference function; This is the cross-entropy difference function.

7. The risk prediction window selection method based on multi-objective collaborative assessment as described in claim 6, characterized in that, Also includes: After determining the theoretically optimal window length according to the selection criteria, the actual modeling process uses a window length that is 1 to 2 time periods shorter than the theoretically optimal window length for model training and updating.

8. A risk prediction window selection device based on multi-objective collaborative assessment, characterized in that, include: The acquisition module is used to acquire the time-series historical data of the target object and divide it into a training subset and a test subset for simulating future predictions in chronological order. The training module is used to set a set of candidate prediction windows containing at least two different time lengths. For each window length in the set, a corresponding binary classification prediction task label is defined based on the length, and a corresponding prediction model is trained using the training subset. All models use the same feature input. An evaluation module is used to evaluate the performance of each of the prediction models on the test subset, wherein the performance evaluation includes at least two-dimensional quantitative metrics. The optimization module is used to select the optimal window length from the candidate window set based on the model performance corresponding to the longest baseline window in the candidate window and according to the preset selection criteria. The selection criterion is: to find the first window length from shortest to longest in which the performance difference between the corresponding model and the benchmark model in multidimensional performance indicators is less than their respective preset thresholds.

9. An electronic device, wherein, The electronic device includes: A processor; and a memory storing computer-executable instructions, which, when executed, cause the processor to perform the method as described in any one of claims 1-7.

10. A computer-readable storage medium, wherein, The computer-readable storage medium stores one or more programs that, when executed by a processor, implement the method of any one of claims 1-7.

Images