Prediction method and device for bank account opening process, storage medium and electronic equipment

Abstract

The invention provides a bank account opening process prediction method and device, a storage medium and electronic equipment, and the method comprises the steps: obtaining all node data of a bank account opening process at a historical moment, obtaining detailed process node data, and carrying out the preliminary screening of the node data, and obtaining initial process node data; evaluating the initial flow node data by adopting multiple algorithms, and screening historical flow node features from the initial flow node data; establishing an initial Patch Mixer model based on a CNN model structure, and training the initial model by using the historical process node features to obtain a target Patch Mixer model; and inputting the target process node features into the target Patch Mixer model to obtain the predicted remaining time of the bank account opening process. According to the method, automatic feature screening and the Patch Mixer model are fused, a prediction result with high accuracy and robustness can be provided, and the method has high universality.

Description

Technical Field

[0001] The present application relates to the field of process management, and in particular, to a method for predicting a bank account opening process, a device for predicting a bank account opening process, a computer-readable storage medium, and an electronic device. Background Art

[0002] As an important process that requires interaction with customers, the accurate prediction of the remaining time can help improve customer satisfaction. Predicting the remaining time can let customers know which node they have reached in the process, which nodes are still to be processed, and the estimated completion time, thereby reducing customers' uncertainty and anxiety. Therefore, it is very important to predict the remaining time of the bank account opening process.

[0003] However, the existing prediction methods have some shortcomings, such as: static prediction models cannot take into account the dynamic changes in the bank account opening process, so the prediction results are inaccurate, the prediction accuracy is not high enough, and there is a large gap between the predicted value and the actual value. Existing prediction methods usually use manual screening features, which increases labor costs and time consumption. In addition, existing prediction methods are usually targeted at specific bank account opening processes, lack predictions in different situations, and cannot be generalized to other bank account opening processes, resulting in inflexible prediction results. Summary of the invention

[0004] The main purpose of the present application is to provide a bank account opening process prediction method, a bank account opening process prediction device, a computer-readable storage medium and an electronic device, so as to at least solve the problems of inaccurate prediction results, low generalization ability and high labor cost of the existing bank account opening process prediction method.

[0005] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a method for predicting a bank account opening process is provided, comprising: obtaining node data of all processes of the bank account opening process at a historical moment to obtain detailed process node data, and preliminarily screening the detailed process node data to obtain initial process node data; using multiple algorithms to evaluate the initial process node data, and screening out historical process node features from the initial process node data; establishing an initial Patch Mixer model based on a CNN model structure, and training the initial Patch Mixer model using the historical process node features to obtain a target Patch Mixer model; inputting the target process node features into the target Patch Mixer model to obtain the predicted remaining time of the bank account opening process, wherein the target process node features are the node data of the processes that have ended in the bank account opening process at the current moment.

[0006] Optionally, multiple algorithms are used to evaluate the initial process node data, and historical process node features are screened out from the initial process node data, including: inputting the initial process node data into each algorithm model in turn to obtain a single importance score of the initial process node data, wherein the single importance score corresponds one-to-one to the algorithm model; determining the corresponding comprehensive importance score of the initial process node data based on all the single importance scores; and determining the initial process node data whose comprehensive importance score is greater than or equal to a threshold score as the historical process node feature.

[0007] Optionally, based on all the single importance scores, the comprehensive importance score of the corresponding initial process node data is determined, including: obtaining multiple score weights, the score weights corresponding one-to-one to the single importance scores; based on all the single importance scores and the corresponding score weights, a weighted sum algorithm is used to calculate the comprehensive importance score of the initial process node data.

[0008] Optionally, the plurality of algorithms include: LightGBM model, XGBoost model, random forest model, Permutation Importance model, recursive feature elimination model and Boruta model.

[0009] Optionally, the historical process node features are used to train the initial Patch Mixer model to obtain a target Patch Mixer model, including: arranging the historical process node features in a chronological sequence, and splitting them into multiple time windows according to the arranged node sequence, each of the time windows including multiple historical process node features, and the historical process node features in two adjacent time windows are partially the same; training the initial Patch Mixer model according to all the time windows and the corresponding historical process node features to obtain the target Patch Mixer model.

[0010] Optionally, the initial Patch Mixer model is trained according to all the time windows and the corresponding historical process node features to obtain the target Patch Mixer model, including: converting the arranged node sequence into Patch embedding to obtain multiple Patches, and the number of the Patches is the same as the number of the time windows; inputting each of the Patches in the feature map, and convolving each of the Patches with a convolution kernel respectively to perform depth convolution on the initial Patch Mixer model; after performing depth convolution on the initial Patch Mixer model, performing point-by-point convolution on the initial Patch Mixer model to train the initial Patch Mixer model to obtain the target Patch Mixer model.

[0011] Optionally, node data of all processes of the bank account opening process at a historical moment is obtained to obtain detailed process node data, and the detailed process node data is preliminarily screened to obtain initial process node data, including: obtaining process nodes, timestamps and process names corresponding to all processes of the bank account opening process at a historical moment; determining a set of the process nodes, the timestamps and the process names corresponding to all the processes as the detailed process node data; and preliminarily screening and merging the detailed process node data to eliminate redundant node data in the detailed process node data to obtain the initial process node data.

[0012] According to another aspect of the present application, a prediction device for a bank account opening process is provided, comprising: an acquisition unit, configured to acquire node data of all processes of the bank account opening process at a historical moment, obtain detailed process node data, and preliminarily screen the detailed process node data to obtain initial process node data; a first processing unit, configured to evaluate the initial process node data using a plurality of algorithms, and screen out historical process node features from the initial process node data; a second processing unit, configured to establish an initial Patch Mixer model based on a CNN model structure, and train the initial Patch Mixer model using the historical process node features to obtain a target Patch Mixer model; and a third processing unit, configured to input the target process node features into the target Patch Mixer model to obtain a predicted remaining time of the bank account opening process, wherein the target process node features are the node data of the processes that have been completed in the bank account opening process at the current moment.

[0013] According to another aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute any one of the methods for predicting the bank account opening process.

[0014] According to another aspect of the present application, an electronic device is provided, comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include a prediction method for executing any one of the bank account opening processes described.

[0015] By applying the technical solution of the present application, the above-mentioned prediction method of the bank account opening process first obtains the node data of all processes of the bank account opening process at the historical moment, obtains the detailed process node data, and preliminarily screens the detailed process node data to obtain the initial process node data; then, a plurality of algorithms are used to evaluate the initial process node data, and the historical process node features are screened out from the initial process node data; then, an initial PatchMixer model is established based on the CNN model structure, and the initial Patch Mixer model is trained using the historical process node features to obtain the target Patch Mixer model; finally, the target process node features are input into the target Patch Mixer model to obtain the predicted remaining time of the bank account opening process, and the target process node features are the node data of the processes that have been completed in the bank account opening process at the current moment. This method innovatively integrates automatic feature screening with the PatchMixer model. The automatic feature screening technology combines the advantages of multiple algorithms to obtain a more representative and robust feature subset and conduct a more comprehensive feature evaluation. The PatchMixer model uses block and hybrid technology to solve the problems of inaccurate prediction results, low generalization ability and high labor costs of existing bank account opening process prediction methods, improve the accuracy and flexibility of predictions, improve the service efficiency and quality of banks, and at the same time, reduce customer uncertainty and anxiety, thereby improving customer satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings constituting part of the present application are used to provide a further understanding of the present application. The exemplary embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0017] Figure 1 A hardware structure block diagram of a mobile terminal for executing a prediction method for a bank account opening process provided in an embodiment of the present application is shown;

[0018] Figure 2A schematic diagram of a process flow of a bank account opening process prediction method provided according to an embodiment of the present application is shown;

[0019] Figure 3 A schematic diagram of a process of an automatic feature screening method provided according to an embodiment of the present application is shown;

[0020] Figure 4 A schematic diagram of a process for training filtered features using a Patch Mixer model according to an embodiment of the present application is shown;

[0021] Figure 5 A schematic diagram of converting an input time series into a Patch embedding using a sliding window method according to an embodiment of the present application is shown;

[0022] Figure 6 A structural block diagram of a device for predicting a bank account opening process provided according to an embodiment of the present application is shown.

[0023] The above drawings include the following reference numerals:

[0024] 102, processor; 104, memory; 106, transmission device; 108, input and output devices. DETAILED DESCRIPTION

[0025] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0026] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0027] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0028] For the convenience of description, some nouns or terms involved in the embodiments of the present application are explained below:

[0029] Automatic feature screening: refers to the use of computer algorithms to automatically screen out valuable features from large amounts of data for further analysis and prediction.

[0030] Feature Importance Analysis: A method used in machine learning to evaluate the impact of each feature (i.e., input variable) on the model's predictive ability.

[0031] LightGBM (Light Gradient Boosting Machine): A distributed high-performance machine learning framework based on gradient boosting.

[0032] XGBoost (Extreme Gradient Boosting): An optimized gradient boosting decision tree (GBDT) algorithm.

[0033] Random Forest: An ensemble learning method.

[0034] Permutation Importance: A model-independent feature importance analysis method.

[0035] Recursive Feature Elimination (RFE): A feature selection method that recursively trains a model and gradually eliminates unimportant features.

[0036] Boruta: A full feature selection method based on random forest.

[0037] Time Series Forecasting: refers to using past time series data to predict future values ​​or states. Time series data is a series of data points arranged in chronological order and can be used to predict future time trends.

[0038] PatchTST: An emerging prediction model based on patch processing of time series.

[0039] PatchMixer: A model specifically designed for time series analysis and forecasting.

[0040] Bank account opening process: refers to the entire process of bank customer account opening, including customer application for account opening, bank review and approval, etc.

[0041] As introduced in the background technology, the static prediction model in the prior art cannot calculate the dynamic changes in the bank account opening process, the prediction results are inaccurate, and the manual screening of features increases labor costs and time consumption. At the same time, the existing methods do not have generalization capabilities and are limited in application scenarios. Although there are some emerging prediction models based on deep learning and reinforcement learning methods, the above models cannot take into account the ability to handle long-term dependencies, model complexity and computing costs, model computing efficiency and flexibility, model interpretability and adaptability, data efficiency and training speed, model stability and prediction accuracy, adjustment and deployment costs, etc. In order to solve the problems of inaccurate prediction results, low generalization ability and high labor costs in the existing bank account opening process prediction methods, the embodiments of the present application provide a bank account opening process prediction method, a bank account opening process prediction device, a computer-readable storage medium and an electronic device.

[0042] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0043] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Taking running on a mobile terminal as an example, Figure 1 1 is a hardware structure block diagram of a mobile terminal of a method for predicting a bank account opening process according to an embodiment of the present invention. Figure 1 As shown, the mobile terminal may include one or more ( Figure 1 Only one is shown in the figure) a processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data, wherein the mobile terminal may also include a transmission device 106 and an input / output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the mobile terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.

[0044] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the prediction method of the bank account opening process in the embodiment of the present invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, the above method is implemented. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories can be connected to the mobile terminal via a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof. The transmission device 106 is used to receive or send data via a network. The above-mentioned specific examples of the network may include a wireless network provided by a communication provider of the mobile terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0045] In this embodiment, a method for predicting a bank account opening process running on a mobile terminal, a computer terminal or a similar computing device is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0046] Figure 2 FIG. 1 is a flow chart of a method for predicting a bank account opening process according to an embodiment of the present application. Figure 2 As shown, the method comprises the following steps:

[0047] Step S201, obtaining node data of all processes of the bank account opening process at a historical moment, obtaining detailed process node data, and preliminarily screening the detailed process node data to obtain initial process node data;

[0048] Specifically, the initial process node data is obtained by Figure 3 The data collection module, basic data statistics module, process node screening and merging module and initialization original feature list module shown are completed, aiming to obtain effective initial process node data and provide a highly relevant and accurate data set for subsequent evaluation of the above node data.

[0049] The above data collection module collects detailed data about the public user process from the production environment. These data include process nodes, timestamps and other relevant information, providing basic data for subsequent analysis.

[0050] The above basic data statistics module performs in-depth statistical analysis on the collected data, including calculating the data distribution of each process node, analyzing the time delay between nodes, and identifying any potential anomalies or trends.

[0051] The above process node screening and merging module combines the analysis of data and the in-depth understanding of business processes to carefully screen and merge process nodes, aiming to simplify the process, eliminate redundancy and highlight key nodes, thereby obtaining a streamlined process that includes events that are critical to predictive modeling.

[0052] The above-mentioned initialization original feature list module uses the above-mentioned selected process events as the initial feature list, which are unique and highly relevant to the prediction model.

[0053] Step S202, using a variety of algorithms to evaluate the initial process node data, and filter out historical process node features from the initial process node data;

[0054] Specifically, Figure 3 Six advanced algorithms are shown for evaluating the importance of each feature, including: LightGBM feature prioritization for calculating gain scores, XGBoost feature prioritization for calculating gain or weight scores, Random Forest feature prioritization for calculating mean square error (MSE), PermutationImportance feature prioritization for calculating mean absolute error (MAE), Recursive feature elimination feature prioritization for calculating mean absolute error (MAE), Boruta feature prioritization for calculating Z-score. The above algorithms set different thresholds to filter features according to business characteristics, provide different angles and methods to measure the contribution of features to model predictions, extract meaningful information from complex data, and select a set of powerful features to build accurate and robust prediction models. This systematic approach combines statistical analysis, business insights, and advanced machine learning techniques to make accurate decisions in modeling the account opening process.

[0055] Among them, multiple algorithms are used to evaluate the above initial process node data, and historical process node features are screened out from the above initial process node data, including the following steps:

[0056] Step S2021, inputting the above-mentioned initial process node data into each algorithm model in sequence to obtain a single importance score of the above-mentioned initial process node data, wherein the above-mentioned single importance score corresponds one-to-one with the above-mentioned algorithm model;

[0057] Specifically, for each algorithm, we calculate the importance score of each feature using the corresponding formulas, including gain, mean square error (MSE), mean absolute error (MAE), and Z-score, which quantify the impact of each feature on the model performance.

[0058] The gain score calculation formula is given by formula (1):

[0059] Gain=∑ j∈splits ΔLoss j (1)

[0060] In the formula, Gain represents the importance score of the feature in the model, ∑j∈∈splits represents the sum of all split nodes j, and the formula will accumulate the contribution of the current feature at different split points, ΔLoss j Represents the change in the loss function of the j-th split node before and after the feature split.

[0061] The mean square error (MSE) calculation formula is given by formula (2):

[0062]

[0063] In the formula, MSE represents the average value of the square difference between the model prediction value and the actual value, y i is the actual value of the i-th sample, is the predicted value of the i-th sample, and n is the number of samples.

[0064] The mean absolute error (MAE) calculation formula is given by formula (3):

[0065]

[0066] In the formula, MAE represents the average value used to measure the absolute difference between the model prediction value and the actual value, y i is the actual value of the i-th sample, is the predicted value of the i-th sample, and nn is the number of samples.

[0067] The Z-score calculation formula is given by formula (4):

[0068]

[0069] In the formula, Z-score is used to evaluate the importance of a feature relative to the significance of "pseudo features" (shadow features), T i represents the importance score of the i-th true feature, T s，i represents the importance score of the i-th “pseudo feature” (shadow feature), and m represents the number of pseudo features.

[0070] Step S2022, determining the comprehensive importance score of the corresponding initial process node data according to all the single importance scores;

[0071] Step S2023, determining the above-mentioned initial process node data whose above-mentioned comprehensive importance score is greater than or equal to the threshold score as the above-mentioned historical process node feature.

[0072] Specifically, recognizing that different algorithms have different feature preferences, an automatic feature selection method based on a fusion algorithm is adopted. By setting different business thresholds, features are automatically selected according to business needs and model objectives, thereby screening out the most relevant features, ensuring that the final feature set is both comprehensive and targeted, and improving the accuracy and efficiency of prediction.

[0073] According to all the above single importance scores, determining the comprehensive importance score of the corresponding above initial process node data includes the following steps:

[0074] Step S301, obtaining a plurality of score weights, wherein the score weights correspond one-to-one to the single importance score;

[0075] Step S302 , based on all the above single importance scores and the corresponding score weights, a weighted sum algorithm is used to calculate the above comprehensive importance score of the above initial process node data.

[0076] Specifically, the comprehensive importance score can more comprehensively evaluate the relative importance of each process node, avoid relying on only one feature to cause large errors in the prediction results, and improve the stability and interpretability of the model.

[0077] Among them, the various algorithms mentioned above include: LightGBM model, XGBoost model, random forest model, permutation importance model, recursive feature elimination model and Boruta model.

[0078] The multiple algorithms may be at least two of the above algorithms, or may be more than the number of the above algorithms. The number of specific algorithms is not specifically limited and may be adjusted according to actual conditions.

[0079] Specifically, the LightGBM model uses a tree-based learning algorithm, which is efficient, fast, and uses less memory. The feature importance analysis methods of the above model mainly include: Split Gain, which is to calculate the information gain brought by the feature when the node in the tree is split. The larger the information gain, the more important the feature is to the model decision; Frequency, which is to count the number of times the feature is used as a split point in all trees. The higher the frequency of occurrence, the more important the feature is to the model; SHAP Values ​​(Shapley Additive exPlanations), which is a feature contribution analysis method based on game theory, can explain the contribution of each feature to each prediction result.

[0080] Specifically, the XGBoost model has the characteristics of high efficiency, accuracy, and flexibility. The above model performs feature importance analysis in the following ways: Weight, which is to count the number of times a feature is used as a split point in all trees. The more times, the more important the feature is; Gain, which is to calculate the average gain brought by the feature when splitting. The greater the gain, the more important the feature is; Cover, which is to calculate the sum of the sample weights used to split the feature. The greater the coverage, the more important the feature is. The above model can also use SHAP Values ​​to explain the contribution of features to the prediction results.

[0081] Specifically, the random forest model improves the accuracy and robustness of the model by constructing multiple decision trees and combining their prediction results. The feature importance analysis methods of the above model include: Mean Decrease Impurity, which is to calculate the importance of the feature based on the Gini impurity or information gain reduced when each feature is split in all trees, and calculate the average impurity reduction value of a feature in all trees; Mean Decrease Accuracy, which is to calculate the importance of the feature by randomly perturbing the value of a feature and then observing the degree of reduction in model accuracy. The more the accuracy decreases, the higher the importance of the feature; Permutation Importance, which is to evaluate the importance of the feature by randomly perturbing the feature value and measuring the change in model performance. The greater the performance decrease, the higher the importance of the feature.

[0082] Among them, Permutation Importance is a model-independent feature importance analysis method. The above method evaluates the importance of a feature by randomly perturbing the value of a feature and then measuring the change in model performance.

[0083] Specifically, the recursive feature elimination algorithm trains a baseline model and sorts the features according to their importance (e.g., absolute value of coefficients or importance in tree models), removes the least important feature (or set of features), and retrains the model using the remaining features until a predetermined number of features or performance criteria is reached.

[0084] Specifically, the model based on the Boruta algorithm evaluates the importance of each original feature by adding noise features (called shadow features).

[0085] Step S203, establishing an initial Patch Mixer model based on the CNN model structure, and using the historical process node features to train the initial Patch Mixer model to obtain a target Patch Mixer model;

[0086] Specifically, the Patch Mixer model training process consists of Figure 4 It is shown that the design of the PatchMixer architecture can capture local and global patterns, and can capture complex patterns and interdependencies in the data, enabling it to handle the complexity and changes that may arise in the bank account opening process. At the same time, the use of the PatchMixer model, through its innovative blocking and mixing technology, significantly improves the accuracy, flexibility, computational efficiency and robustness of time series prediction, and can effectively handle long-term dependencies, nonlinear relationships and multivariate data.

[0087] The core idea of ​​the Patch Mixer model is to divide time series data into small fragments (patches) and model and predict time series data by mixing these fragments.

[0088] The above historical process node features are used to train the above initial Patch Mixer model to obtain the target Patch Mixer model, including the following steps:

[0089] Step S2031, arranging the above historical process node features in a time sequence, and splitting the arranged node sequence into multiple time windows, each of the above time windows includes multiple of the above historical process node features, and the above historical process node features in two adjacent time windows are partially the same;

[0090] Step S2032: training the initial Patch Mixer model according to all the time windows and the corresponding historical process node features to obtain the target Patch Mixer model.

[0091] Specifically, this method uses a sliding window approach to convert the input time series into Patch embedding. Figure 5A clear conversion process is given. The horizontal axis of the figure represents time, and the vertical axis represents time consumption. The above method divides the time series data into multiple windows of equal length. Figure 5 Four windows (window 1, window 2, window 3, window 4) are shown. These windows are similar to sliding fields of view, and each window captures the value of a time period. In each window, the originally continuous time series data is divided into smaller nodes, which are the input information that needs to be processed in the window, and the last two nodes in the window are the same as the first two nodes in the adjacent window, that is, A and B, C and D, E and F in the figure are the same nodes. This method fragments the original time series for further embedding and processing, and the windows in the figure are continuous on the time axis. This sliding window method helps to preserve temporal dependencies and information flow.

[0092] The initial Patch Mixer model is trained according to all the time windows and the corresponding historical process node features to obtain the target Patch Mixer model, including the following steps:

[0093] Step S401, converting the arranged node sequence into patch embedding to obtain multiple patches, where the number of the patches is the same as the number of the time windows;

[0094] Specifically, Patch Embedding is a patch-based representation that decomposes sequential time series data into smaller structured fragments, similar to image patches. The extension to a 2D format introduces spatial considerations, consistent with the nature of image data, and supports the utilization of convolution operations, which are well suited to capture local patterns and global relationships in time series.

[0095] Step S402, input each of the above patches in the feature map, and convolve each of the above patches with a convolution kernel respectively, so as to perform deep convolution on the above initial Patch Mixer model;

[0096] Specifically, this method uses a special grouped convolution where the number of groups is equal to the number of patches. In order to expand the receptive field of the model, a larger convolution kernel size is used. Each patch in the input feature map is convolved with a convolution kernel. This operation generates multiple feature maps, each corresponding to a specific patch, and then these feature maps are sequentially connected to create an output feature map with multiple channels. Deep convolution effectively uses the same grouped convolution kernel for patches sharing the same spatial position, which allows the model to capture potential periodic patterns within the temporal patch.

[0097] Step S403, after performing depth convolution on the initial Patch Mixer model, performing point-by-point convolution on the initial Patch Mixer model to train the initial Patch Mixer model to obtain the target Patch Mixer model.

[0098] Specifically, since the deep convolution operation cannot effectively capture the feature correlation between patches, a point-by-point convolution is performed after the deep convolution. Through the point-by-point convolution, the temporal interaction between patches is realized, thereby enhancing the model's ability to capture the correlation of input features.

[0099] Specifically, local position information represents the temporal order of time series data, which is of great significance. The position-free encoding strategy of the Patch Mixer model bypasses the limitation that the self-attention layer in the traditional Transformer model architecture cannot inherently retain this position information through the local perception ability of the convolutional layer and the local feature extraction of the sliding window. In order to enhance the temporal context of the time series input, it is necessary to add position encoding to represent the temporal order. This method uses the CNN structure, which inherently has permutation variance, eliminating the need to use position embedding in the model.

[0100] In some instances, Figure 4 The model training process of using the Patch Mixer model to train the filtered features to predict the remaining time is given, including the following steps:

[0101] The time series data is passed into the model as input features and normalized. The feature mean of each sample is made 0 and the variance is 1 through the instance normalization method. The normalized time series features are converted into multiple nodes, namely, patches, as the input of the subsequent neural network.

[0102] Each node is feature extracted through deep convolution. The deep convolution method convolves each channel independently to capture the local spatial relationship within each node. Then, the GELU activation function and BatchNorm are applied to further improve the stability of training.

[0103] Point-by-point convolution is a 1*1 convolution, which extracts global information across nodes by combining features between different channels. Then, GELU activation function and BatchNorm (batch normalization) are also applied to further improve the stability of training.

[0104] The features extracted by the convolutional layer are subsequently input into the fully connected layer to generate the features required for prediction. The GELU activation function is introduced in the fully connected layer to ensure that the model can handle complex time series patterns. Finally, all features are flattened into a one-dimensional vector and applied to the output layer to generate the remaining time prediction value.

[0105] Step S204, input the target process node feature into the target Patch Mixer model to obtain the predicted remaining time of the bank account opening process. The target process node feature is the node data of the completed process in the bank account opening process at the current moment.

[0106] Specifically, the existing prediction methods mainly include: statistical methods based on historical data, methods based on machine learning, and methods based on deep learning.

[0107] The above statistical methods include applying the autoregression model (AR) to use past values ​​to predict future values, applying the moving average model (MA) to use past prediction errors to predict future values, and applying ARMA and ARIMA models. This type of model is a hybrid model that combines autoregression and moving average. ARIMA adds a difference operation to handle non-stationary data. This type of method is suitable for processing stable time series with more significant linear relationships, but performs poorly in nonlinear, non-stationary or long-term prediction scenarios.

[0108] The above machine learning-based methods specifically include linear regression models, which are simple linear models that can be used for basic time series forecasting, as well as support vector regression models (SVR) and decision trees and random forests, among which tree-based models can handle nonlinear relationships. This method can handle more complex relationships and data features, but cannot capture long-term dependencies, and usually has multiple hyperparameters that need to be adjusted, increasing the complexity and computational cost of model tuning.

[0109] The above-mentioned deep learning-based methods specifically include RNN (recurrent neural network), especially LSTM (long short-term memory network) and GRU (gated recurrent unit), which are specially designed to process sequence data, and CNN (convolutional neural network) can be used to extract local patterns in time series data. In addition, in recent years, the Transformer architecture and its variants, such as Temporal Fusion Transformer, PatchTST, etc. have performed well in time series prediction. Among them, PatchTST is more focused on using the Transformer's self-attention mechanism to capture long-term dependencies, but it has problems such as poor flexibility and high adjustment and deployment costs.

[0110] There is also a method based on reinforcement learning, which optimizes the prediction process by building an environment based on time series prediction and training intelligent agents. For example, Deep Q-Network (DQN) or Proximal Policy Optimization (PPO) can be applied to this field, especially for optimizing predictions in dynamic environments, but it cannot fully achieve the comprehensive technical advantages of this application.

[0111] The Patch Mixer used in this application focuses on processing time series data through hybrid technology, which has higher flexibility and simplicity. The automatic feature screening process enhances the prediction model by identifying and selecting the most relevant input features. By eliminating redundant or unimportant features, the model focuses on the most critical factors for prediction. By integrating the automatic feature screening technology and the Patch Mixer model, the prediction model can achieve higher accuracy, thereby more accurately predicting the remaining time of the bank account opening process.

[0112] The process remaining time prediction method proposed in this application is highly universal. It is not only applicable to the bank account opening process, but can also be expanded to be applied to the remaining time prediction of other long business processes, providing a powerful prediction tool for various process management.

[0113] In some examples, obtaining node data of all processes of the bank account opening process at a historical moment, obtaining detailed process node data, and preliminarily screening the above detailed process node data to obtain initial process node data includes the following steps:

[0114] Step S501, obtaining the process nodes, timestamps and process names corresponding to all processes of the bank account opening process at a historical moment;

[0115] Step S502, determining a set of the process nodes, the timestamps and the process names corresponding to all the processes as the detailed process node data;

[0116] Step S503 , preliminarily screening and merging the above detailed process node data to eliminate redundant node data in the above detailed process node data to obtain the above initial process node data.

[0117] Specifically, it aims to obtain effective initial process node data and provide a highly relevant and accurate data set for subsequent evaluation of the above node data to obtain historical process node characteristics.

[0118] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0119] The present application embodiment also provides a prediction device for a bank account opening process. It should be noted that the prediction device for the bank account opening process of the present application embodiment can be used to execute the prediction method for the bank account opening process provided by the present application embodiment. The device is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions have been made and will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware of a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware is also possible and conceived.

[0120] The following is an introduction to the bank account opening process prediction device provided in the embodiment of the present application.

[0121] Figure 6 Schematic diagram of a device for predicting a bank account opening process according to an embodiment of the present application. Figure 6 As shown, the device includes: an acquisition unit 10, a first processing unit 20, a second processing unit 30 and a third processing unit 40, the acquisition unit 10 is used to acquire the node data of all processes of the bank account opening process at a historical moment, obtain detailed process node data, and preliminarily screen the above detailed process node data to obtain initial process node data; the first processing unit 20 is used to evaluate the above initial process node data using multiple algorithms, and screen out historical process node features from the above initial process node data; the second processing unit 30 is used to establish an initial Patch Mixer model based on the CNN model structure, and use the above historical process node features to train the above initial Patch Mixer model to obtain a target Patch Mixer model; the third processing unit 40 is used to input the target process node features into the above target Patch Mixer model to obtain the predicted remaining time of the above bank account opening process, and the above target process node features are the node data of the process that has been completed in the above bank account opening process at the current moment.

[0122] The prediction device for the above-mentioned bank account opening process of the present application includes: an acquisition unit, a first processing unit, a second processing unit and a third processing unit, the acquisition unit is used to acquire the node data of all processes of the bank account opening process at a historical moment, obtain detailed process node data, and preliminarily screen the detailed process node data to obtain initial process node data; the first processing unit is used to evaluate the initial process node data using multiple algorithms, and screen out historical process node features from the initial process node data; the second processing unit is used to establish an initial Patch Mixer model based on the CNN model structure, and train the initial Patch Mixer model using historical process node features to obtain a target Patch Mixer model; the third processing unit is used to input the target process node features into the target Patch Mixer model to obtain the predicted remaining time of the bank account opening process, and the target process node features are the node data of the process that has ended in the bank account opening process at the current moment. The device combines multiple feature selection algorithms based on different criteria to obtain a more representative and robust feature subset, thereby improving model performance, stability and interpretability while reducing the risk of overfitting and having stronger adaptability. At the same time, combined with the PatchMixer model, it uses its innovative blocking and mixing technology to achieve accurate prediction of the remaining time of the bank account opening process. This not only greatly improves the degree of automation of the prediction process, but also fully considers the dynamic nature of the process, thereby achieving a dual improvement in prediction efficiency and accuracy, and solving the problems of inaccurate prediction results, low generalization ability and high labor costs in existing bank account opening process prediction methods.

[0123] In some examples, the first processing unit includes a first processing module, a second processing module and a third processing module. The first processing module is used to input the above-mentioned initial process node data into each algorithm model in turn to obtain a single importance score of the above-mentioned initial process node data, and the above-mentioned single importance score corresponds to the above-mentioned algorithm model one by one; the second processing module is used to determine the corresponding comprehensive importance score of the above-mentioned initial process node data according to all the above-mentioned single importance scores; the third processing module is used to determine the above-mentioned initial process node data whose comprehensive importance score is greater than or equal to the threshold score as the above-mentioned historical process node feature. In this way, the historical process node features that are unique and highly relevant to the prediction model can be efficiently screened out to improve the accuracy and efficiency of the prediction.

[0124] In some instances, the second processing module also includes a first processing submodule and a second processing submodule, the first processing submodule is used to obtain multiple score weights, the above score weights correspond to the above single importance scores one by one; the second processing submodule is used to calculate the above comprehensive importance score of the above initial process node data using a weighted summation algorithm based on all the above single importance scores and the corresponding above score weights. In this way, by combining the advantages of multiple algorithms, the dependence on specific data patterns can be reduced, the stability of the results and the predictive performance of the model can be improved, and at the same time, a variety of information can be integrated to conduct a more comprehensive feature evaluation, reduce the risk of bias and overfitting, enhance the model's interpretability, and have stronger adaptability to cope with various data features and problem types.

[0125] In some examples, the above algorithms include LightGBM model, XGBoost model, random forest model, Permutation Importance model, recursive feature elimination model and Boruta model. These algorithms have different advantages, and can complete automatic feature screening by setting thresholds, reducing labor costs and time consumption.

[0126] In some instances, the second processing unit includes a fourth processing module and a fifth processing module, the fourth processing module is used to arrange the above historical process node features in a time sequence, and split the node sequence after arrangement to obtain multiple time windows, each of the above time windows includes multiple historical process node features, and the above historical process node features in two adjacent time windows are partially the same; the fifth processing module is used to train the above initial Patch Mixer model according to all the above time windows and the corresponding historical process node features to obtain the above target Patch Mixer model. This can eliminate the limitation of the traditional Transformer model that needs to introduce position encoding to represent the time sequence, has the ability to handle long-term dependencies, and improves the calculation efficiency and flexibility of the model.

[0127] In some instances, the second processing unit further includes a sixth processing module, a seventh processing module, and an eighth processing module. The sixth processing module is used to convert the above-arranged node sequence into a patch embedding to obtain multiple patches, and the number of the above patches is the same as the number of the above time windows; the seventh processing module is used to input each of the above patches in the feature map, and convolve each of the above patches with a convolution kernel respectively to perform deep convolution on the above initial Patch Mixer model; the eighth processing module is used to perform point-by-point convolution on the above initial Patch Mixer model after deep convolution on the above initial Patch Mixer model, so as to train the above initial Patch Mixer model and obtain the above target Patch Mixer model. In this way, through innovative block and mixing technologies, key issues such as long-term dependence, computational complexity, model flexibility, nonlinear relationship capture, and multivariate data processing are solved.

[0128] In some examples, the acquisition unit includes a first acquisition module, a second acquisition module and a third acquisition module. The first acquisition module is used to acquire the process nodes, timestamps and process names corresponding to all processes of the bank account opening process at historical moments; the second acquisition module is used to determine the set of the process nodes, timestamps and process names corresponding to all the above processes as the above detailed process node data; the third acquisition module is used to preliminarily screen and merge the above detailed process node data to eliminate redundant node data in the above detailed process node data to obtain the above initial process node data. In this way, the prediction model is enhanced by identifying and selecting the most relevant input features. At the same time, by eliminating redundant or unimportant features, the model can focus on the most critical factors for prediction.

[0129] The bank account opening process prediction device includes a processor and a memory. The acquisition unit and the like are stored in the memory as program units, and the processor executes the program units stored in the memory to implement corresponding functions. The modules are all located in the same processor; or, the modules are located in different processors in any combination.

[0130] The processor includes a kernel, which retrieves the corresponding program unit from the memory. One or more kernels can be set, and the kernel parameters can be adjusted to solve the problems of inaccurate prediction results, low generalization ability and high labor cost of the existing bank account opening process prediction method.

[0131] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0132] An embodiment of the present invention provides a computer-readable storage medium, which includes a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute the prediction method of the bank account opening process.

[0133] Specifically, the prediction methods for the bank account opening process include:

[0134] Step S201, obtaining node data of all processes of the bank account opening process at a historical moment, obtaining detailed process node data, and preliminarily screening the detailed process node data to obtain initial process node data;

[0135] Step S202, using a variety of algorithms to evaluate the initial process node data, and filter out historical process node features from the initial process node data;

[0136] Step S203, establishing an initial Patch Mixer model based on the CNN model structure, and using the historical process node features to train the initial Patch Mixer model to obtain a target Patch Mixer model;

[0137] Step S204, input the target process node feature into the target Patch Mixer model to obtain the predicted remaining time of the bank account opening process. The target process node feature is the node data of the completed process in the bank account opening process at the current moment.

[0138] Optionally, multiple algorithms are used to evaluate the above-mentioned initial process node data, and historical process node features are screened out from the above-mentioned initial process node data, including: inputting the above-mentioned initial process node data into each algorithm model in turn to obtain a single importance score of the above-mentioned initial process node data, and the above-mentioned single importance score corresponds one-to-one to the above-mentioned algorithm model; determining the corresponding comprehensive importance score of the above-mentioned initial process node data based on all the above-mentioned single importance scores; and determining the above-mentioned initial process node data whose above-mentioned comprehensive importance score is greater than or equal to the threshold score as the above-mentioned historical process node feature.

[0139] Optionally, based on all the above single importance scores, the corresponding comprehensive importance score of the above initial process node data is determined, including: obtaining multiple score weights, the above score weights correspond one-to-one to the above single importance scores; based on all the above single importance scores and the corresponding above score weights, a weighted sum algorithm is used to calculate the above comprehensive importance score of the above initial process node data.

[0140] Optionally, the plurality of the above algorithms include: LightGBM model, XGBoost model, random forest model, Permutation Importance model, recursive feature elimination model and Boruta model.

[0141] Optionally, the above-mentioned historical process node features are used to train the above-mentioned initial Patch Mixer model to obtain the target Patch Mixer model, including: arranging the above-mentioned historical process node features in a chronological sequence, and splitting them into multiple time windows according to the arranged node sequence, each of the above-mentioned time windows includes multiple of the above-mentioned historical process node features, and the above-mentioned historical process node features in two adjacent time windows are partially the same; the above-mentioned initial Patch Mixer model is trained according to all the above-mentioned time windows and the corresponding historical process node features to obtain the above-mentioned target Patch Mixer model.

[0142] Optionally, the initial Patch Mixer model is trained according to all the above time windows and the corresponding historical process node features to obtain the target Patch Mixer model, including: converting the arranged node sequence into Patch embedding to obtain multiple Patches, and the number of the Patches is the same as the number of the time windows; inputting each of the above Patches in the feature map, and convolving each of the above Patches with a convolution kernel respectively to perform depth convolution on the initial Patch Mixer model; after performing depth convolution on the initial Patch Mixer model, performing point-by-point convolution on the initial Patch Mixer model to train the initial Patch Mixer model to obtain the target Patch Mixer model.

[0143] Optionally, node data of all processes of the bank account opening process at a historical moment is obtained to obtain detailed process node data, and the above detailed process node data is preliminarily screened to obtain initial process node data, including: obtaining process nodes, timestamps and process names corresponding to all processes of the bank account opening process at a historical moment; determining the set of the above process nodes, the above timestamps and the above process names corresponding to all the above processes as the above detailed process node data; preliminarily screening and merging the above detailed process node data to eliminate redundant node data in the above detailed process node data to obtain the above initial process node data.

[0144] An embodiment of the present invention provides a processor, which is used to run a program, wherein the program executes the prediction method of the bank account opening process when running.

[0145] Step S201, obtaining node data of all processes of the bank account opening process at a historical moment, obtaining detailed process node data, and preliminarily screening the detailed process node data to obtain initial process node data;

[0146] Step S202, using a variety of algorithms to evaluate the initial process node data, and filter out historical process node features from the initial process node data;

[0147] Step S203, establishing an initial Patch Mixer model based on the CNN model structure, and using the historical process node features to train the initial Patch Mixer model to obtain a target Patch Mixer model;

[0148] Step S204, input the target process node feature into the target Patch Mixer model to obtain the predicted remaining time of the bank account opening process. The target process node feature is the node data of the completed process in the bank account opening process at the current moment.

[0149] An embodiment of the present invention provides a device, the device including a processor, a memory, and a program stored in the memory and executable on the processor, and when the processor executes the program, at least the following steps are implemented:

[0150] Step S201, obtaining node data of all processes of the bank account opening process at a historical moment, obtaining detailed process node data, and preliminarily screening the detailed process node data to obtain initial process node data;

[0151] Step S202, using a variety of algorithms to evaluate the initial process node data, and filter out historical process node features from the initial process node data;

[0152] Step S203, establishing an initial Patch Mixer model based on the CNN model structure, and using the historical process node features to train the initial Patch Mixer model to obtain a target Patch Mixer model;

[0153] Step S204, input the target process node feature into the target Patch Mixer model to obtain the predicted remaining time of the bank account opening process. The target process node feature is the node data of the completed process in the bank account opening process at the current moment.

[0154] The devices in this article can be servers, PCs, PADs, mobile phones, etc.

[0155] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program for initializing at least the following method steps:

[0156] Step S201, obtaining node data of all processes of the bank account opening process at a historical moment, obtaining detailed process node data, and preliminarily screening the detailed process node data to obtain initial process node data;

[0157] Step S202, using a variety of algorithms to evaluate the initial process node data, and filter out historical process node features from the initial process node data;

[0158] Step S203, establishing an initial Patch Mixer model based on the CNN model structure, and using the historical process node features to train the initial Patch Mixer model to obtain a target Patch Mixer model;

[0159] Step S204, input the target process node feature into the target Patch Mixer model to obtain the predicted remaining time of the bank account opening process. The target process node feature is the node data of the completed process in the bank account opening process at the current moment.

[0160] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0161] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0162] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0163] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0164] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0165] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0166] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0167] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0168] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0169] From the above description, it can be seen that the above embodiments of the present application achieve the following technical effects:

[0170] 1) The prediction method of the above-mentioned bank account opening process of the present application first obtains the node data of all processes of the bank account opening process at the historical moment, obtains the detailed process node data, and preliminarily screens the detailed process node data to obtain the initial process node data; then, a variety of algorithms are used to evaluate the initial process node data, and the historical process node features are screened out from the initial process node data; then, an initial Patch Mixer model is established based on the CNN model structure, and the initial Patch Mixer model is trained using the historical process node features to obtain a target Patch Mixer model; finally, the target process node features are input into the target Patch Mixer model to obtain the predicted remaining time of the bank account opening process, and the target process node features are the node data of the process that has ended in the bank account opening process at the current moment. This method innovatively integrates automatic feature screening with the PatchMixer model. The automatic feature screening technology combines the advantages of multiple algorithms to obtain a more representative and robust feature subset and conduct a more comprehensive feature evaluation. The Patch Mixer model uses blocking and mixing technology to solve the problems of inaccurate prediction results, low generalization ability and high labor costs of existing bank account opening process prediction methods, improve the accuracy and flexibility of predictions, improve the service efficiency and quality of banks, and at the same time reduce customer uncertainty and anxiety, thereby improving customer satisfaction.

[0171] 2) The prediction device for the above-mentioned bank account opening process of the present application includes: an acquisition unit, a first processing unit, a second processing unit and a third processing unit. The acquisition unit is used to acquire the node data of all processes of the bank account opening process at a historical moment, obtain detailed process node data, and preliminarily screen the detailed process node data to obtain initial process node data; the first processing unit is used to evaluate the initial process node data using multiple algorithms, and screen out historical process node features from the initial process node data; the second processing unit is used to establish an initial Patch Mixer model based on the CNN model structure, and train the initial Patch Mixer model using historical process node features to obtain a target Patch Mixer model; the third processing unit is used to input the target process node features into the target Patch Mixer model to obtain the predicted remaining time of the bank account opening process, and the target process node features are the node data of the process that has ended in the bank account opening process at the current moment. The device combines multiple feature selection algorithms based on different criteria to obtain a more representative and robust feature subset, thereby improving model performance, stability and interpretability while reducing the risk of overfitting and having stronger adaptability. At the same time, combined with the PatchMixer model, it uses its innovative blocking and mixing technology to achieve accurate prediction of the remaining time of the bank account opening process. This not only greatly improves the degree of automation of the prediction process, but also fully considers the dynamic nature of the process, thereby achieving a dual improvement in prediction efficiency and accuracy, and solving the problems of inaccurate prediction results, low generalization ability and high labor costs in existing bank account opening process prediction methods.

[0172] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for predicting a bank account opening process, characterized in that: include: Obtaining node data of all processes of the bank account opening process at a historical moment, obtaining detailed process node data, and preliminarily screening the detailed process node data to obtain initial process node data; Using multiple algorithms to evaluate the initial process node data, and screening out historical process node features from the initial process node data; An initial Patch Mixer model is established based on the CNN model structure, and the historical process node features are used to train the initial Patch Mixer model to obtain a target Patch Mixer model; The target process node feature is input into the target Patch Mixer model to obtain the predicted remaining time of the bank account opening process, and the target process node feature is the node data of the process that has been completed in the bank account opening process at the current moment.

2. The method according to claim 1, characterized in that A variety of algorithms are used to evaluate the initial process node data and filter out historical process node features from the initial process node data, including: Inputting the initial process node data into each algorithm model in turn to obtain a single importance score of the initial process node data, wherein the single importance score corresponds to the algorithm model one by one; Determine the comprehensive importance score of the corresponding initial process node data according to all the single importance scores; The initial process node data whose comprehensive importance score is greater than or equal to a threshold score is determined as the historical process node feature.

3. The method according to claim 2, characterized in that Determining the comprehensive importance score of the corresponding initial process node data according to all the single importance scores includes: Acquire a plurality of score weights, wherein the score weights correspond one-to-one to the single importance score; According to all the single importance scores and the corresponding score weights, a weighted sum algorithm is adopted to calculate the comprehensive importance score of the initial process node data.

4. The method according to claim 1, characterized in that: The various algorithms include: LightGBM model, XGBoost model, random forest model, permutation importance model, recursive feature elimination model and Boruta model.

5. The method according to claim 1, characterized in that: The initial Patch Mixer model is trained using the historical process node features to obtain a target Patch Mixer model, including: Arrange the historical process node features in a time sequence, and split the arranged node sequence to obtain multiple time windows, each of the time windows includes multiple historical process node features, and the historical process node features in two adjacent time windows are partially the same; The initial Patch Mixer model is trained according to all the time windows and the corresponding historical process node features to obtain the target Patch Mixer model.

6. The method according to claim 5, characterized in that The initial Patch Mixer model is trained according to all the time windows and the corresponding historical process node features to obtain the target Patch Mixer model, including: Convert the arranged node sequence into patch embedding to obtain multiple patches, where the number of the patches is the same as the number of the time windows; Input each of the patches in the feature map, and convolve each of the patches with a convolution kernel to perform deep convolution on the initial Patch Mixer model; After performing depth convolution on the initial Patch Mixer model, performing point-by-point convolution on the initial Patch Mixer model to train the initial Patch Mixer model to obtain the target Patch Mixer model.

7. The method according to any one of claims 1 to 6, characterized in that: Obtain node data of all processes of the bank account opening process at a historical moment, obtain detailed process node data, and preliminarily screen the detailed process node data to obtain initial process node data, including: Get the process nodes, timestamps and process names corresponding to all processes of the bank account opening process at historical moments; Determine a set of the process nodes, the timestamps, and the process names corresponding to all the processes as the detailed process node data; The detailed process node data are preliminarily screened and merged to eliminate redundant node data in the detailed process node data to obtain the initial process node data.

8. A bank account opening process prediction device, characterized in that: include: An acquisition unit, used to acquire node data of all processes of the bank account opening process at a historical moment, obtain detailed process node data, and preliminarily screen the detailed process node data to obtain initial process node data; A first processing unit, configured to evaluate the initial process node data using a plurality of algorithms, and filter out historical process node features from the initial process node data; A second processing unit is used to establish an initial Patch Mixer model based on the CNN model structure, and train the initial Patch Mixer model using the historical process node features to obtain a target Patch Mixer model; The third processing unit is used to input the target process node feature into the target Patch Mixer model to obtain the predicted remaining time of the bank account opening process, wherein the target process node feature is the node data of the process that has been completed in the bank account opening process at the current moment.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the bank account opening process prediction method according to any one of claims 1 to 7.

10. An electronic device, characterized in that: include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include a prediction method for executing the bank account opening process described in any one of claims 1 to 7.

Images