Tail box balance prediction method of bank automatic teller machine and cash replenishing strategy
By building an integrated model based on random forest and XGBoost algorithm, accurately predicting the trunk balance of bank ATMs is solved, and the problem of bank ATM inventory cash management relies on subjective prediction in the existing technology is improved, and the efficiency and accuracy of capital flows are improved.
Patent Information
- Application Number
- CN202411991446.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-31
- Publication Date
- 2025-05-30
AI Technical Summary
The existing bank ATM cash inventories relies on the subjective predictions of managers, and the lack of precise data support and the assistance of scientific models, resulting in redundancy or lack of cash in inventory, affecting the efficient use of funds.
By obtaining the time series data of ATM trunk balances, building a data set, and based on a model combining the random forest algorithm and XGBoost algorithm, it predicts the ATM trunk balances of different groups of types to achieve accurate prediction of the ATM trunk balance.
It improves the accuracy of forecasting the bank's ATM trunk balance, optimizes the capital flow planning, reduces the risk of redundant or lack of cash in stocks, and improves the efficiency of bank business operations.
Smart Images
Figure CN120069156A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the financial field and relates to a method for predicting the balance of the cash box of a bank automated teller machine and a cash replenishment strategy. Background Art
[0002] With the booming development of the digital economy, while changing the shopping methods of consumers, the way of capital flow has become more diversified. The emergence of digital currency and online payment has brought about a huge change in the bank's cash business. The demand for cash in stock is affected by various factors. Allocating according to the subjective prediction of managers is likely to cause problems of cash redundancy or shortage, resulting in waste of human and material resources and being unfavorable to the efficient use of funds. However, there is little relevant research, and the large amount of data stored in banks has not been taken seriously. It is urgent to accurately predict the demand for cash in bank ATMs.
[0003] Although mobile payment and online payment have had a certain impact on the bank's cash business, for areas with relatively inconvenient transportation, low network penetration, and groups who are not very familiar with online payment, the cash business still plays an indispensable role. However, the existing bank cash inventory management relies on the subjective prediction of managers, lacking accurate data support and the assistance of scientific models, resulting in problems of cash redundancy or shortage, affecting the efficient use of funds. At the same time, the shortage of cash may cause customers to be unable to withdraw money when needed, while the redundancy of cash causes waste of resources and affects the satisfaction of customer service. Summary of the Invention
[0004] To solve the above problems, the technical solution adopted by the present invention is: a method for predicting the balance of the cash box of a bank automated teller machine, including the following steps:
[0005] S1: Obtain the time series data of the ATM cash box balance and construct a data set;
[0006] S2: Based on the data set, divide the group types to which the automated teller machines belong;
[0007] S3: Based on the data after dividing the group types, respectively construct an ATM cash box balance prediction model for predicting the balance of the bank ATM cash box of different group types by combining the random forest algorithm and the XGBoost algorithm;
[0008] S4: Use the training set data to train the ATM cash box balance prediction models of different group types to obtain the trained different ATM cash box balance prediction models;
[0009] S5: Based on the trained different ATM cash box balance prediction models, realize the prediction of the balance of the bank ATM cash box.
[0010] Further: The process of classifying the group types to which ATMs belong based on the data set is as follows:
[0011] Based on the time series data of the balance of the ATM cash box, use a box plot to visualize the time series data of the balance of the ATM cash box to determine the distribution characteristics of the data;
[0012] Based on the distribution characteristics of the box plot, determine a single threshold standard baseline, and compare the average value of the box plot of the historical data of the balance of each ATM cash box with the single threshold standard baseline to achieve the classification of the group types to which the ATMs belong.
[0013] Further: The process of comparing the average value of the box plot of the historical data of the balance of each ATM cash box with the single threshold standard baseline to achieve the classification of the group types to which the ATMs belong is as follows:
[0014] When the average value of the box plot of the historical data of the balance of the ATM cash box ≥ the single threshold standard baseline, the ATM is in an active transaction community;
[0015] When the average value of the box plot of the historical data of the balance of the ATM cash box < the single threshold standard baseline, the ATM is in a cold transaction community.
[0016] Further: The ATM cash box balance prediction model uses the random forest algorithm to perform regression analysis on the following features including the cash box balance, 30-day rolling average, 20-day rolling average, 10-day rolling average, and 5-day rolling average for the target variable d+1 cash box balance.
[0017] According to the importance of the influence of each feature on the regression result, rank the above features in descending order;
[0018] Select the above 5, 4, 3, 2, and 1 features in turn to construct an XGBoost regression prediction model as the ATM cash box balance prediction model.
[0019] A device for predicting the balance of the cash box of a bank ATM includes:
[0020] A construction module: used to obtain the time series data of the balance of the ATM cash box and construct a data set;
[0021] A classification module: used to classify the group types to which the ATMs belong based on the data set;
[0022] A prediction model establishment module: Based on the data after classifying the group types, construct an ATM cash box balance prediction model that combines the random forest algorithm and the XGBoost algorithm and is used to predict the balance of the bank ATM cash box of different group types;
[0023] Training model: For the training set data, train the ATM cash box balance prediction models for different group types to obtain different trained ATM cash box balance prediction models;
[0024] Prediction module: Based on the different trained ATM cash box balance prediction models, realize the prediction of the cash box balance of bank ATMs.
[0025] A readable storage medium stores program modules, and the program modules can be run in a processor to implement the method described in any one of the above.
[0026] According to the cash replenishment strategy of the cash box balance prediction method of a bank automated teller machine described in any one of the above, based on the prediction result of the cash box balance of the bank ATM and the cash replenishment method, realize the prediction of the final cash replenishment amount of the automated teller machine.
[0027] Further: The cash replenishment method is as follows:
[0028] Taking the predicted result demand of the cash box balance of the bank ATM and increasing it by X% or taking the fixed cash reserve amount of the bank and subtracting the predicted inventory cash amount and decreasing it by X% as the final cash replenishment strategy.
[0029] The cash box balance prediction method and cash replenishment strategy of a bank automated teller machine provided by the present invention have the following advantages: Based on deep cooperation with third-party financial service institutions, closely fitting the actual operation status of ATMs in commercial bank branches, aiming to solve the problem of the management of the cash in stock of existing bank ATMs. Deeply analyzing the historical operation data of ATMs, through accurately mining and extracting the internal characteristics of the data, fully considering the data distribution characteristics, realizing the accurate classification of the communities of ATMs. On this basis, an integrated model based on the random forest and XGBoost algorithms is proposed and compared with the ARMA model, random forest model, and XGBoost model. The results show that the integrated model has better prediction effects and can well improve the accuracy of fund prediction.
[0030] Based on the suggestions of banking experts, the prediction results can further form a cash replenishment strategy, thereby optimizing the fund flow plan, reducing the risk of redundant or missing cash in stock, and improving the operation efficiency of banking services. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required to be used in the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.
[0032] Figure 1 is the flowchart of the present invention;
[0033] Figure 2 It is a box plot of the tail box balance distribution of the present invention;
[0034] Figure 3 It is the flow chart of the integrated model of the present invention;
[0035] Figure 4 This is a schematic diagram of the banknote adding strategy of the present invention;
[0036] Figure 5 is a box plot of the monthly distribution of the tail box balance of six representative ATMs in the transaction-depleted community of the present invention, wherein (a) ATM 1, (b) ATM 2, (c) ATM 3, (d) ATM 4, (e) ATM 5, and (f) ATM 6;
[0037] Figure 6 It is a box plot of the monthly distribution of the tail box balance of six representative ATMs in the transaction active community of the present invention; wherein (a) ATM 7, (b) ATM 8, (c) ATM 9, (d) ATM 10, (e) ATM 11, (f) ATM 12;
[0038] Figure 7 It is a deserted community for transactions in the present invention: a line graph of prediction errors of different characteristic variable models;
[0039] Figure 8 This is a line graph of prediction errors of different feature variable models for active trading communities of the present invention. DETAILED DESCRIPTION
[0040] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0041] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present invention and its application or use. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0042] Figure 1 is a flow chart of the present invention;
[0043] A method for predicting the balance of a bank automatic teller machine includes the following steps:
[0044] S1: Obtain the time series data of the ATM cash box balance and construct a data set;
[0045] S2: Based on the data set, divide the group types to which the ATMs belong;
[0046] S3: Based on the data after dividing the group types, construct an ATM cash box balance prediction model that combines the random forest algorithm and the XGBoost algorithm and is used to predict the bank ATM cash box balance for different group types;
[0047] S4: Use the training set data to train the ATM cash box balance prediction models for different group types to obtain the trained different ATM cash box balance prediction models;
[0048] S5: Based on the trained different ATM cash box balance prediction models, realize the prediction of the bank ATM cash box balance.
[0049] The steps S1 / S2 / S3 / S4 / S5 are executed in sequence;
[0050] Figure 2 It is the box plot of the cash box balance distribution of the present invention;
[0051] Furthermore: The process of dividing the group types to which the ATMs belong based on the data set is as follows:
[0052] Based on the transaction records of multiple teller numbers included in the ATM cash box balance time series data, use the box plot to visualize the ATM cash box balance time series data to determine the distribution characteristics of the data;
[0053] Based on the distribution characteristics of the box plot, determine a single threshold standard baseline, and compare the average value of the box plot of the historical cash box balance data of each ATM with the single threshold standard baseline to realize the division of the group types to which the ATMs belong;
[0054] The single threshold standard baseline is determined by the following method:
[0055] By adding and subtracting 10% of the total average value of all ATM boxes, the upper limit and the lower limit of the single threshold standard baseline used are determined, that is, Figure 2 the two dotted lines in;
[0056] Within the upper limit and the lower limit, the initial single threshold standard baseline is selected as the total average value - 10%, with a step width of 2%, and the standard baseline + 2% is cycled until the single threshold standard baseline is equal to the total average value + 10%. The final single threshold standard baseline is the maximum standard baseline that can divide the ATM boxes in the figure into two groups most evenly in this search method.
[0057] The process of comparing the average value of the box plot of the historical data of the balance in the cash box of each ATM with the single-threshold standard baseline to achieve the classification of the group types to which the ATMs belong is as follows:
[0058] When the average value of the box plot of the historical data of the balance in the cash box of the ATM ≥ the single-threshold standard baseline, the ATM is in an active trading community;
[0059] When the average value of the box plot of the historical data of the balance in the cash box of the ATM < the single-threshold standard baseline, the ATM is in a cold trading community.
[0060] Furthermore: The prediction effect of the bank ATM cash box balance prediction model is evaluated by comparing the predicted objective function value, i.e., the predicted value, with the true value, and making comparisons by grouping (active trading community, cold trading community) and by model (bank ATM cash box balance prediction model for active trading community, bank ATM cash box balance prediction model for cold trading community).
[0061] Figure 3 It is the integrated model flowchart of the present invention;
[0062] The ATM cash box balance prediction model uses a random forest model to perform regression analysis on the following features, including the cash box balance, 30-day rolling mean, 20-day rolling mean, 10-day rolling mean, and 5-day rolling mean, for the target variable of the cash box balance at d+1.
[0063] According to the importance degree of the influence of each feature on the regression result, the above features are sorted in descending order;
[0064] Select the above 5, 4, 3, 2, and 1 features in sequence to construct an XGBoost regression prediction model as the ATM cash box balance prediction model.
[0065] A device for predicting the balance of the cash box of a bank automated teller machine includes:
[0066] A construction module: used to obtain the time series data of the ATM cash box balance and construct a data set;
[0067] A classification module: used to classify the group types to which the automated teller machines belong based on the data set;
[0068] A prediction model establishment module: based on the data after classifying the group types, respectively construct an ATM cash box balance prediction model that combines the random forest algorithm and the XGBoost algorithm and is used to predict the balance of the bank ATM cash box for different group types;
[0069] Training model: For the training set data, train the ATM cash box balance prediction models for different group types to obtain the trained different ATM cash box balance prediction models;
[0070] Prediction module: Based on the trained different ATM cash box balance prediction models, realize the prediction of the cash box balance of bank ATMs.
[0071] A readable storage medium stores program modules, and the program modules can run in a processor to implement any of the above methods.
[0072] According to the cash replenishment strategy of the cash box balance prediction method of a bank automated teller machine as described in any one of the above, based on the prediction result of the cash box balance of the bank ATM and based on the cash replenishment method, realize the prediction of the final cash replenishment amount of the automated teller machine.
[0073] Further, the cash replenishment method is as follows:
[0074] Taking the predicted result demand of the cash box balance of the bank ATM increased by X% or the fixed cash reserve minus the predicted inventory cash decreased by X% as the final cash replenishment strategy.
[0075] Figure 4 This is the schematic diagram of the cash replenishment strategy of the present invention;
[0076] The process of realizing the prediction of the final cash replenishment amount of the automated teller machine based on the prediction result of the cash box balance of the bank ATM and based on the cash replenishment strategy is as follows:
[0077] Taking the predicted result demand of the cash box balance of the bank ATM increased by X% or the fixed cash reserve minus the predicted inventory cash decreased by X% as the final cash replenishment strategy. The X% is 5%;
[0078] Embodiment 1:
[0079] S1: Data collection: Collect the inventory cash data of commercial bank automated teller machines, including transaction records of multiple teller numbers. Preprocess the data to remove missing data and abnormal data. Normalize the data, change the data in the "date" column from "20160201" to the date format of "2016-02-01", and remove redundant information such as duplicate characters from the data in the "teller number" column. For the data in the cash box balance column, calculate the mean value as new data each time by moving down one unit according to the specifications of 30, 20, 10, and 5, and add "Predict_label" as the cash box balance of day+1.
[0080] S2: Group type classification: Considering that the tail box balance corresponding to different teller numbers is limited by the region, in economically developed and high-consumption areas, transactions are more frequent and the tail box balance is generally lower. For economically underdeveloped and relatively low-consumption areas, transactions are less frequent.
[0081] The balance in the last box is generally high. In addition to geographical restrictions, the youthfulness of the residents in the region also has an impact. Generally speaking, communities with a large concentration of young people have strong consumption power, while communities with a large number of elderly people have weaker consumption power. The data can be divided into communities with low transaction volume and communities with active transaction volume. Grouping steps: Randomly select data from 10 teller numbers and use box plots to visualize these data to determine the distribution characteristics of the data. Preliminary division criteria are set based on the box plots to divide the data into communities with low transaction volume and communities with active transaction volume. The division criteria are adjusted by 10% up and down to optimize the division of communities.
[0082] S3: Community verification: Randomly select 6 teller numbers from each community and draw a monthly distribution box plot. Observe whether the usage patterns in the same community are similar to ensure the rationality of community division. Finally, 2.2×10 5 is the grouping standard.
[0083] Figure 5 is a box plot of the monthly distribution of the tail box balance of six representative ATMs in the transaction-depleted community of the present invention, wherein (a) ATM 1, (b) ATM 2, (c) ATM 3, (d) ATM 4, (e) ATM 5, and (f) ATM 6;
[0084] Figure 6 It is a box plot of the monthly distribution of the tail box balance of six representative ATMs in the transaction active community of the present invention; wherein (a) ATM 7, (b) ATM 8, (c) ATM 9, (d) ATM 10, (e) ATM 11, (f) ATM 12;
[0085] 2. Establishment of ATM tail box balance prediction model and evaluation of prediction effect
[0086] 1. Establishment of ATM tail box balance prediction model: The data of each community is divided into 60% training set, 20% validation set, and 20% test set, and random forest and XGBoost prediction models are established respectively.
[0087] 2. Model evaluation: Compare the predicted objective function value with the true value, calculate the mean square error and mean square root error, R 2 Evaluate the degree of model fit. Calculate the MSE, RMSE, and R of the six teller numbers in groups. 2 The mean is used as the standard for evaluating the prediction effect of the two groups of models. The models are compared by group and model to evaluate the prediction effect of the models.
[0088] 3. Optimization of the ATM tail box balance prediction model: Combine the random forest algorithm with the XGBoost algorithm, select different numbers of features according to the order of feature importance in the random forest to build the XGBoost regression prediction model, that is, according to the order of feature importance in the random forest, gradually reduce the features to build the XGBoost regression prediction model. In order to eliminate the influence of experimental data on the model and accidental factors, the data still uses 6 teller numbers extracted from the above two communities, a total of 12 teller numbers, and selects each feature to build a corresponding model group to calculate the error mean to evaluate the model prediction effect. Finally, it was found that the integrated model of four variables (tail box balance, "20-day rolling mean", "10-day rolling mean", and "5-day rolling mean") has the best prediction effect.
[0089] 3. Implementation of the Strategy of Adding More Notes
[0090] 1. Model prediction: Select the ATM tail box balance prediction model trained with the best hyperparameter configuration to predict the data.
[0091] 2. Developing a strategy for adding more notes: Based on the forecast results and the experience of banking experts, two strategies for adding more notes are developed:
[0092] (1) The predicted demand is increased by 5% to form a cash increase strategy. (2) The fixed cash reserve of each ATM minus the predicted tail box balance is decreased by 5% to form a cash increase strategy.
[0093] 4. Monitoring and Adjustment
[0094] After implementing the cash addition, continuously monitor the cash usage of ATMs and adjust the cash addition strategy in a timely manner based on actual usage and forecast accuracy.
[0095] Figure 7 It is a deserted community for transactions in the present invention: a line graph of prediction errors of different characteristic variable models;
[0096] Figure 8 This is a line graph of prediction errors of different feature variable models for active trading communities of the present invention.
[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for predicting the balance of a bank automatic teller machine, characterized in that: The following steps are involved: S1: Obtain the time series data of the ATM tail box balance and construct a data set; S2: Based on the data set, the group types to which the ATMs belong are divided; S3: Based on the data of group type division, an ATM tail box balance prediction model based on a combination of random forest algorithm and XGBoost algorithm is constructed to predict the bank ATM tail box balance of different group types; S4: training set data, training the ATM tail box balance prediction model for different groups of types, and obtaining trained different ATM tail box balance prediction models; S5: Based on the trained different ATM tail box balance prediction models, the tail box balance of the bank ATM is predicted.
2. The method for predicting the tail box balance of a bank automatic teller machine according to claim 1, characterized in that: The process of dividing the group types to which the ATMs belong based on the data set is as follows: Based on the ATM tail box balance time series data, the box plot is used to visualize the ATM tail box balance time series data to determine the distribution characteristics of the data; Based on the distribution characteristics of the box plot, a single threshold standard baseline is determined, and the average box plot of the historical data of the tail box balance of each ATM is compared with the single threshold standard baseline to realize the classification of the group type to which the ATMs belong.
3. The method for predicting the balance in the tail box of a bank automatic teller machine according to claim 2, characterized in that: The process of comparing the average box plot of the historical data of the tail box balance of each ATM machine with the single threshold standard baseline to realize the classification of the group type to which the ATM belongs is as follows: When the average value of the box figure of the ATM machine's tail box balance historical data is ≥ the single threshold standard baseline, the ATM is in an active transaction community; When the average value of the box chart of the ATM's tail box balance historical data is less than the single threshold standard baseline, the ATM is in a community with low transaction volume.
4. The method for predicting the tail box balance of a bank automatic teller machine according to claim 1, characterized in that: The ATM tail box balance prediction model uses the random forest algorithm to perform regression analysis on the target variable d+1 tail box balance on the following features including tail box balance, 30-day rolling mean, 20-day rolling mean, 10-day rolling mean and 5-day rolling mean. According to the importance of each feature on the regression results, the above features are ranked from high to low; The above 5, 4, 3, 2, and 1 features are selected respectively to build the XGBoost regression prediction model as the ATM tail box balance prediction model.
5. A device for predicting the balance of a tail box of a bank automatic teller machine, characterized in that: include: Construction module: used to obtain the ATM tail box balance time series data and build a data set; Division module: used to divide the group types to which the ATMs belong based on the data set; Prediction model building module: Based on the data of group type classification, an ATM tail box balance prediction model based on the combination of random forest algorithm and XGBoost algorithm is constructed to predict the bank ATM tail box balance of different group types; Training model: used for training set data, training ATM tail box balance prediction models for different groups of types, and obtaining trained different ATM tail box balance prediction models; Prediction module: used to predict the tail box balance of bank ATMs based on different trained ATM tail box balance prediction models.
6. A readable storage medium storing a program module, characterized in that: The program module is executed in a processor to implement the method according to any one of claims 1 to 4.
7. A banknote adding strategy for a method for predicting the balance of a tail box of a bank automatic teller machine according to any one of claims 1 to 5, characterized in that: Based on the prediction results of the bank ATM tail box balance and the banknote adding method, the final banknote adding amount of the ATM is predicted.
8. A method for adding banknotes to an automatic teller machine according to claim 7, characterized in that: The method of adding banknotes is as follows: The final cash replenishment strategy is to increase the predicted demand of the bank ATM's tail box balance by X% or to reduce the fixed cash reserve amount minus the predicted cash inventory by X%.