Banking business peak period prediction method and system based on real-time computing platform

Abstract

The invention provides a bank business peak period prediction method and system based on a real-time computing platform, and the method comprises the steps: S1, collecting historical transaction data and external factor data from a bank database in real time, and carrying out the cleaning and preprocessing of the collected data; s2, extracting the preprocessed time, historical window and external factor features, and carrying out feature coding and engineering optimization processing on the features; s3, constructing a machine learning model, taking the encoded time, historical window and external factor characteristics as input to train the model, adjusting hyper-parameters through cross validation, and evaluating the performance of the model; s4, receiving real-time data, extracting input features, inputting the real-time data into the trained model for reasoning, and outputting a prediction result; and S5, dynamically adjusting the resource configuration of the bank according to the prediction result, and reversely optimizing the model by monitoring the actual operation condition in real time. According to the method, system resources are dynamically adjusted through the real-time prediction result, and the pressure of business peaks on hardware and human resources is relieved.

Description

Technical Field

[0001] The present invention relates to the field of computer information technology, and in particular to a banking business peak period prediction method and system based on a real-time computing platform. Background Art

[0002] With the rapid development of banking business, real-time data processing and prediction are becoming more and more important in the banking industry. At present, the prediction of banking business peak period faces the following main problems: 1. Limitations of traditional methods: Batch processing methods cannot meet real-time requirements, and data collection lags make it difficult for prediction results to reflect business changes in a timely manner. Secondly, rule-based prediction methods cannot adapt to complex nonlinear business models. Therefore, a system that can collect, process and analyze data in real time is needed to predict business peak periods, help banks optimize resource allocation and improve customer service quality. Summary of the invention

[0003] In view of the technical problems existing in the prior art, the present invention provides a banking business peak forecasting method and system based on a real-time computing platform. The present invention realizes low-latency data collection and distribution through OGG and Kafka, combines FlinkSQL for real-time computing and data stream processing, combines random forest, XGBoost and other models, and utilizes time series and external factor characteristics to improve the forecasting accuracy. The system resources are dynamically adjusted through real-time forecasting results to alleviate the pressure of business peak on hardware and human resources.

[0004] According to a first aspect of the present invention, the present invention provides a banking business peak period prediction method based on a real-time computing platform, comprising the following steps:

[0005] Step S1. Collect historical transaction data and external factor data from the bank database in real time, and clean and pre-process the collected data;

[0006] Step S2. extracting the preprocessed time, historical window and external factor features, and performing feature coding and engineering optimization processing on the features;

[0007] Step S3. Build a machine learning model, train the model using the encoded time, historical window and external factor features as input, adjust hyperparameters through cross-validation, and evaluate model performance;

[0008] Step S4. Receive real-time data, extract input features, input the real-time data into the trained model for inference, and output the prediction results;

[0009] Step S5. Dynamically adjust the bank's resource allocation according to the prediction results, and reversely optimize the model by real-time monitoring of actual operations.

[0010] Based on the above technical solution, the present invention can also make the following improvements.

[0011] Optionally, the historical transaction data includes transaction type, amount, timestamp, and transaction channel; the external factor data includes weather data and holiday information.

[0012] Optionally, the data cleaning and preprocessing include: missing value processing, outlier processing and data normalization.

[0013] Optionally, the time feature is represented by the hourly peak value during business hours; the historical window feature is represented by the transaction volume and trend in the past hours; and the external factor feature includes: weather and statutory holiday identification.

[0014] Optionally, extracting the pre-processed time, historical window and external factor features, and performing feature coding and engineering optimization processing on the features includes:

[0015] The time feature is encoded using one-hot encoding, the historical window feature uses a sliding window to generate cumulative transaction volume, and the external factor feature is processed by discretization or standardization.

[0016] Optionally, in constructing the machine learning model, a machine learning model in supervised learning is adopted, including a random forest model or a gradient boosting decision tree, and a time series analysis method is adopted, including an ARIMA model to capture the trend and seasonal characteristics of the time series.

[0017] Optionally, the step of constructing a machine learning model, taking the encoded time, historical window, and external factor features as input to train the model, adjusting hyperparameters through cross-validation, and evaluating model performance includes:

[0018] The dataset is split into 80% for training and 20% for testing, ensuring that the data in the test set does not participate in model training;

[0019] Use the training set to train the model. During the model training process, use the cross-validation method to adjust the model's hyperparameters.

[0020] The test set is used to evaluate the model performance, and the error indicator is used to measure the prediction effect of the model.

[0021] Optionally, the real-time data is input into a trained model for inference, and the output prediction result includes:

[0022] Features consistent with the training process are extracted from the real-time features, and the extracted features are input into the trained machine learning model to obtain the predicted value of the business peak period, and whether the business peak period is likely to occur is determined based on the predicted value; the predicted value is the transaction volume in a future period of time.

[0023] Optionally, the bank's resource allocation is dynamically adjusted according to the prediction results, and the reverse optimization model includes:

[0024] According to the predicted business peak period, the bank rationally arranges resources, including: when the business peak period is predicted to come, increase service windows or customer service staff in advance to reduce customers' waiting time;

[0025] Dynamically adjust the configuration of system servers based on business forecast results, including increasing system server resources to cope with concurrent requests when business peaks arrive;

[0026] Comparing the forecast results with the actual business volume helps managers judge the accuracy of the model forecast. If the forecast deviates greatly from the actual situation, the forecast accuracy can be further improved by adjusting the model or optimizing features.

[0027] According to a second aspect of the present invention, there is provided a banking business peak period prediction system based on a real-time computing platform, comprising:

[0028] The data collection module is used to collect historical transaction data and external factor data from the bank database in real time, and clean and pre-process the collected data;

[0029] A data processing module is used to extract the pre-processed time, historical window and external factor features, and perform feature coding and engineering optimization processing on the features;

[0030] The model training and prediction module is used to build a machine learning model, train the model using the encoded time, historical window, and external factor features as input, adjust hyperparameters through cross-validation, evaluate model performance, receive real-time data, extract input features, input real-time data into the trained model for inference, and output prediction results;

[0031] The result application module is used to dynamically adjust the bank's resource allocation according to the prediction results and reversely optimize the model by real-time monitoring of actual operations.

[0032] Technical effects and advantages of the present invention:

[0033] The present invention provides a banking business peak forecasting method and system based on a real-time computing platform, which ensures low-latency data processing by combining real-time data collection and processing with machine learning; combines historical transaction characteristics and external factor characteristics to achieve accurate prediction through machine learning algorithms. Feature engineering optimization is adopted to analyze data in multiple dimensions through time characteristics (hours, days of the week), historical window characteristics (sliding windows, lag characteristics) and external factor characteristics (weather, holidays) to improve prediction accuracy. Real-time monitoring of actual business performance is carried out and compared with the forecast results to form a closed-loop optimization mechanism. The system architecture has expansion capabilities and can dynamically adjust data collection and processing strategies according to business volume to achieve real-time prediction of banking business peak periods. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a schematic diagram of a banking business peak period prediction system based on a real-time computing platform provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. 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.

[0036] It is understandable that, based on the defects in the background technology, the embodiment of the present invention proposes a banking business peak period prediction method based on a real-time computing platform, comprising the following steps:

[0037] Step S1. Collect historical transaction data and external factor data from the bank database in real time, and then clean and pre-process the collected data;

[0038] Specifically, business data from MySQL or Oracle databases is collected through OGG, and the data is written into the Kafka message queue. Historical transaction data is obtained from the bank's business system, including transaction type, amount, timestamp, transaction channel, etc.; external factor data is obtained, including weather data, holiday information, etc. The combination of historical transaction data and external factor data provides rich contextual support for feature engineering. For example, the combination of holiday information and historical windows can help generate holiday-specific business trend features.

[0039] Data cleaning refers to the process of preprocessing and converting data to meet the needs of data analysis and data mining and to improve data quality. This process usually includes operations such as deleting duplicate information, correcting errors, resolving missing values, and deleting outliers to ensure that subsequent data analysis and data mining are more accurate and effective. The purpose of data cleaning is to improve the accuracy, consistency, and completeness of data, ensure that the data can truly reflect the business situation, and provide a reliable basis for data analysis and decision-making. Through data cleaning, the accuracy of data can be improved, the impact of errors and outliers can be reduced, the consistency of data can be enhanced, the problem of data islands can be solved, the reliability of data analysis results can be guaranteed, and business efficiency can be improved.

[0040] Data cleaning and preprocessing include:

[0041] Missing value processing: Missing values ​​are processed by mean interpolation or filling with adjacent data to ensure data integrity.

[0042] Outlier processing: Detect and remove outliers in the data, such as a sudden increase or decrease in trading volume during normal business hours.

[0043] Data normalization: Normalize the data to the same range to eliminate the dimensional differences between features and avoid some features from causing unnecessary bias in prediction.

[0044] The cleaned and preprocessed data provides a high-quality and complete data source for feature engineering.

[0045] Step S2. extracting the preprocessed time, historical window and external factor features, and performing feature coding and engineering optimization processing on the features;

[0046] In this embodiment, Flink SQL is used to perform real-time feature encoding and engineering optimization processing on the data in Kafka.

[0047] Time features are expressed as: hour, day of the week, month, etc. (e.g. hourly peaks during business hours);

[0048] The historical window features are represented by the trading volume and trend of the past 1 hour and 3 hours;

[0049] External factors include: weather (precipitation, temperature, etc.) and holidays (statutory holiday signs).

[0050] It should be noted that time features are a key part of feature engineering, which can help the model identify periodic fluctuations and trends in business volume, including basic time features, periodic time features, and composite time features; among which, the basic time features mainly include:

[0051] Hourly feature: Decompose the transaction timestamp into hourly features. Usually, business peaks occur in specific hours, so the time feature is very important.

[0052] Extract the hour information (0-23) from the transaction timestamp. For example, banking volume may peak between 9am and 11am or between 2pm and 4pm.

[0053] Application scenario: Analyze the specific time periods of business peaks and predict the concentrated transaction periods of the day.

[0054] Coding method: directly use digital coding (such as 9 represents 9 o'clock) or convert it into periodic features through sine and cosine transform (suitable for machine learning models).

[0055] Day of the week feature: Different weekdays and weekends have different impacts on business volume. The day of the week can be one-hot encoded or marked in integer form.

[0056] The day of the week that the extracted transaction occurred (0 for Sunday, 1 for Monday, and so on).

[0057] Application scenario: There may be significant differences in business volume between weekdays and weekends, with weekend business volume typically lower and Monday or Friday possibly higher.

[0058] Encoding method: One-Hot Encoding or digital encoding can be used.

[0059] Date Features:

[0060] Extracting specific date information (such as the beginning, middle, and end of the month) is particularly useful for identifying the impact of month-end bill payments or salary payments on business volume.

[0061] Application scenario: Identify business peak characteristics at the beginning or end of the month.

[0062] Coding method: The date can be discretized into three types of features: "beginning of the month", "middle of the month" and "end of the month".

[0063] The seasonality feature is represented by extracting the season (such as spring, summer, autumn, winter) from the transaction time. For example, the Spring Festival or National Day holiday may have a significant impact on the business volume.

[0064] Application scenario: predicting business changes caused by seasonal fluctuations.

[0065] Coding method: Use a binary variable to identify a specific season (e.g. 1 for spring, 0 for non-spring).

[0066] Peak trading period mark:

[0067] Mark specific peak periods based on historical banking business data, such as morning peak period, lunch break period, and evening peak period.

[0068] Application scenario: predicting the business volume at different time periods in a day.

[0069] Coding method: Map the trading time period (such as 9:00-11:00) to a categorical variable.

[0070] The composite time features include: time combination features. Composite features are generated by combining date and time. For example, "2024-11-11 10:30:00" can be decomposed into the "Monday-Morning Peak" feature;

[0071] Application scenario: Capturing the impact of interactions between different time dimensions.

[0072] Time trend features: Create trend variables (such as time series that start from 0 and increase incrementally) to help the model identify long-term trends in business volume growth or decline.

[0073] Application scenario: long-term business forecasting, such as the growth forecast of the total transaction volume for the whole year.

[0074] Time characteristics can intuitively reflect the cyclical fluctuations of banking business volume (such as daily peaks, weekly peaks, and monthly peaks). When the above time characteristics are combined with external factors (such as holidays and weather), the prediction model's ability to identify business peaks can be significantly improved.

[0075] The historical window feature is represented as follows: using a sliding window to generate trading volume features, such as the trading volume and trend of the past hour and the past three hours, which can help the model learn short-term trends.

[0076] The historical window features include lag features, which are used to generate the trading volume of the previous hour and the previous three hours. The lag features help the model capture the correlation in the time series.

[0077] The external factor characteristics are represented by: weather (rainfall, temperature, etc.) and holidays (statutory holiday marks). The external factor characteristics can improve the accuracy of the model's prediction of business peak periods.

[0078] Weather characteristics: Weather conditions (e.g., temperature, rainfall, etc.) may affect banking business volume. For example, customers may reduce offline business volume on rainy days.

[0079] Holiday characteristics: The business volume on holidays is significantly different from that on ordinary days. A binary variable can be set to mark whether a day is a holiday.

[0080] The connection between characteristics and business scenarios: Transaction type and transaction channel directly reflect business attributes, while time and weather characteristics indirectly affect customer behavior and business volume.

[0081] The feature encoding and optimization of time, historical window and external factor features specifically include:

[0082] The time feature uses One-Hot Encoding, the historical window feature uses a sliding window to generate cumulative transaction volume, and the external factor feature is processed by discretization or standardization.

[0083] The above time, historical window and external factor features are encoded and used as input for model training in step S3.

[0084] Step S3. Build a machine learning model, train the model using the encoded time, historical window and external factor features as input, adjust hyperparameters through cross-validation, and evaluate model performance;

[0085] In this embodiment, machine learning models in supervised learning can be used, such as Random Forest or XGBoost, which can better handle complex nonlinear relationships. Time series analysis methods, such as ARIMA model, can also be selected to capture the trend and seasonal characteristics of time series.

[0086] The model training and evaluation steps include:

[0087] Data splitting: Split the dataset into 80% for training and 20% for testing, ensuring that the data in the test set does not participate in model training to evaluate the generalization ability of the model;

[0088] Model training: Use the training set to train the model. During the model training process, you can use the cross-validation method to adjust the model's hyperparameters, such as the number of trees in the random forest and the learning rate in XGBoost;

[0089] Model evaluation: Use the test set to evaluate the model performance, and use common error indicators such as mean square error (MSE) and root mean square error (RMSE) to measure the prediction effect of the model.

[0090] Step S4. Receive real-time data, extract input features, input the real-time data into the trained model for inference, and output the prediction results;

[0091] In this implementation step, the machine learning model trained in step S3 is used for real-time reasoning in step S4. During the reasoning process, the feature extraction process of real-time data must be consistent with that during model training to ensure that the model can correctly reason and output accurate prediction results. The following are the specific feature categories that need to be extracted and processed:

[0092] Time features are used to capture the time periodicity of the business; they include:

[0093] Hour feature: the hour information of the current data (such as 9 means 9 am).

[0094] Function: To predict the business volume at a certain hour of the day.

[0095] Encoding: Direct numerical or sine-cosine transforms to express periodicity.

[0096] Day of the week features:

[0097] The current day of the week (e.g. 1 for Monday).

[0098] Function: Analyze the impact of weekdays and weekends on business volume.

[0099] Date Features:

[0100] Whether it is the beginning, middle or end of the month.

[0101] Purpose: Analyze the peak of bill payments at the end of the month.

[0102] Holiday logo:

[0103] Whether the current date is a statutory holiday or the day before a holiday.

[0104] Function: To capture the impact of holidays on business.

[0105] The historical window feature is used to capture short-term business trends, including:

[0106] Sliding window features:

[0107] Cumulative trading volume in the past 1 hour, 3 hours, and 6 hours.

[0108] Function: Capture short-term trend changes and determine whether business volume is increasing or decreasing. Lag characteristics:

[0109] Trading volume for the past 1 hour or 3 hours.

[0110] Purpose: Helps the model identify lag effects in time series.

[0111] Transaction characteristics, used to directly reflect the core data of banking business; including: Transaction type:

[0112] Such as classified information of deposits, withdrawals, transfers, etc.

[0113] Purpose: Identify the contribution of different types of transactions to business volume.

[0114] Transaction amount:

[0115] The single or cumulative transaction amount at the current moment.

[0116] Function: Analyze the impact of transaction amount on business pressure.

[0117] Trading channels:

[0118] The source of the transaction, such as offline branches, ATMs, self-service machines or mobile banking.

[0119] Function: Identify the business distribution patterns of different channels.

[0120] External factor characteristics are used to eliminate the indirect impact of the external environment on banking business; including:

[0121] Weather characteristics: current temperature, rainfall, etc.

[0122] Function: Analyze the inhibitory effect of bad weather on business volume (e.g. customers reduce offline business volume on rainy days). Holiday characteristics:

[0123] Holiday category (such as Spring Festival, National Day).

[0124] Function: To capture the impact of important holidays on business volume fluctuations.

[0125] Real-time statistics features:

[0126] Dynamic statistical features generated in combination with real-time data.

[0127] Current concurrent transaction volume:

[0128] The number of concurrent transactions in a certain period of time.

[0129] Function: Reflect the system load condition.

[0130] Current transaction peak mark:

[0131] Whether the real-time transaction volume exceeds the historical peak value in a certain period of time in the past.

[0132] Purpose: Quickly mark potential peak hours.

[0133] Combination and interaction features

[0134] Capture more complex influence patterns through feature combinations.

[0135] Time and weather interaction features:

[0136] The interactive impact of weather at different time periods on business volume (such as the morning rush hour on rainy days).

[0137] Transaction type and time characteristics:

[0138] The business volume patterns of different transaction types in a specific time period.

[0139] Feature extraction example If the current time is 10:00 on November 11, 2024, the extracted features may include:

[0140] Time characteristics: hours: 10

[0141] Day of the week: 1 (Monday)

[0142] Is it a holiday: 0 (No)

[0143] History Window Features:

[0144] Transaction volume in the past hour: 500

[0145] Cumulative transaction volume in the past 3 hours: 1200

[0146] Transaction Features:

[0147] The main transaction type in the current period: deposits (accounting for 60%)

[0148] Cumulative transaction amount: 2 million yuan

[0149] External factors characteristics:

[0150] Weather: Sunny, temperature 15℃, no rain

[0151] Real-time statistics features:

[0152] Number of concurrent transactions: 50

[0153] After the model training is completed, the real-time prediction phase can be entered. Real-time prediction requires that the newly arrived data be passed through the trained model to quickly make predictions. The real-time data is input into the trained model for inference, and the output prediction results include:

[0154] Feature extraction: Extract features consistent with the training process from real-time data, such as hourly features, sliding window features, etc., to ensure that real-time data and training data are consistent in feature space;

[0155] Model reasoning: The extracted features are input into the trained machine learning model to obtain the predicted value of the business peak period. The predicted result is usually the transaction volume in the next hour or a period of time in the future. The predicted value can be used to determine whether a business peak period is likely to occur.

[0156] Step S5. Dynamically adjust the bank's resource allocation according to the prediction results, and reversely optimize the model by real-time monitoring of actual operations.

[0157] In this embodiment, the generated prediction results are used in specific business scenarios, such as dynamically adjusting the bank's resource allocation, monitoring the actual operation status in real time, and reversely optimizing the model to improve the prediction effect.

[0158] The prediction results have a wide range of application scenarios in actual business, mainly reflected in the following aspects:

[0159] Resource optimization: Based on the predicted business peak period, banks can arrange resources reasonably. For example, when the business peak period is predicted to come, service windows or customer service staff can be added in advance to reduce customers' waiting time.

[0160] Equipment configuration: The system server configuration can be dynamically adjusted based on the results of business forecasts. When business peaks, the system server resources can be increased to cope with concurrent requests.

[0161] Operation monitoring: Compare the forecast results with the actual business volume to help managers judge the accuracy of the model forecast. If the forecast deviates greatly from the actual situation, the forecast accuracy can be further improved by adjusting the model or optimizing features.

[0162] In summary, the embodiment of the present invention ensures low-latency data processing by integrating OGG, Kafka, Flink SQL, PostgreSQL and ClickHouse technologies; uses historical window features and time features, combined with machine learning models to improve the accuracy of prediction. It realizes real-time prediction of the peak period of banking business, which can significantly improve the business operation efficiency and customer service level of the bank.

[0163] According to a second aspect of the present invention, a banking business peak period prediction system based on a real-time computing platform is provided. The system adopts real-time computing technology, including data collection, data processing and data analysis, to achieve real-time prediction of banking business peak periods. Figure 1 As shown, the system comprises:

[0164] The data collection module is used to collect historical transaction data and external factor data from the bank database in real time, and clean and pre-process the collected data;

[0165] A data processing module is used to extract the pre-processed time, historical window and external factor features, and perform feature coding and engineering optimization processing on the features;

[0166] Model training and prediction module, which is used to build machine learning models, train the models by taking the encoded time, historical windows and external factor features as input, adjust hyperparameters through cross-validation, and evaluate model performance; and receive real-time data, extract input features, input real-time data into the trained model for inference, and output prediction results;

[0167] The result application module is used to dynamically adjust the bank's resource allocation according to the prediction results and reversely optimize the model by real-time monitoring of actual operations.

[0168] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0169] 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 make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A banking business peak period prediction method based on a real-time computing platform, characterized in that: The following steps are involved: Step S1. Collect historical transaction data and external factor data from the bank database in real time, and clean and pre-process the collected data; Step S2. extracting the preprocessed time, historical window and external factor features, and performing feature coding and engineering optimization processing on the features; Step S3. Build a machine learning model, train the model using the encoded time, historical window and external factor features as input, adjust hyperparameters through cross-validation, and evaluate model performance; Step S4. Receive real-time data, extract input features, input the real-time data into the trained model for inference, and output the prediction results; Step S5. Dynamically adjust the bank's resource allocation according to the prediction results, and reversely optimize the model by real-time monitoring of actual operations.

2. A banking business peak period prediction method based on a real-time computing platform according to claim 1, characterized in that: The historical transaction data includes transaction type, amount, timestamp, and transaction channel; the external factor data includes weather data and holiday information.

3. A banking business peak period prediction method based on a real-time computing platform according to claim 2, characterized in that: The data cleaning and preprocessing include: missing value processing, outlier processing and data normalization.

4. The method for predicting the peak period of banking business based on a real-time computing platform according to claim 1, characterized in that: The time feature is represented by the hourly peak value during business hours; the historical window feature is represented by the transaction volume and trend in the past hour; The external factor characteristics include: weather and statutory holiday identification.

5. The method for predicting the peak period of banking business based on a real-time computing platform according to claim 1, characterized in that: The extracting of pre-processed time, historical window and external factor features, and performing feature coding and engineering optimization processing on the features include: The time feature is encoded using one-hot encoding, the historical window feature uses a sliding window to generate cumulative transaction volume, and the external factor feature is processed by discretization or standardization.

6. A banking business peak period prediction method based on a real-time computing platform according to claim 1, characterized in that: In constructing the machine learning model, a machine learning model in supervised learning is adopted, including a random forest model or a gradient boosting decision tree, and a time series analysis method, including an ARIMA model, is adopted to capture the trend and seasonal characteristics of the time series.

7. A banking business peak period prediction method based on a real-time computing platform according to claim 1, characterized in that: The machine learning model is constructed, the encoded time, historical window and external factor features are used as input to train the model, the hyperparameters are adjusted through cross-validation, and the model performance is evaluated including: The dataset is split into 80% for training and 20% for testing, ensuring that the data in the test set does not participate in model training; Use the training set to train the model. During the model training process, use the cross-validation method to adjust the model's hyperparameters. The test set is used to evaluate the model performance, and the error indicator is used to measure the prediction effect of the model.

8. The method for predicting the peak period of banking business based on a real-time computing platform according to claim 1, characterized in that: The real-time data is input into the trained model for inference, and the output prediction results include: Features consistent with the training process are extracted from the real-time features, and the extracted features are input into the trained machine learning model to obtain the predicted value of the business peak period, and whether the business peak period is likely to occur is determined based on the predicted value; the predicted value is the transaction volume in a future period of time.

9. A banking business peak period prediction method based on a real-time computing platform according to claim 1, characterized in that: The reverse optimization model includes: dynamically adjusting the bank's resource allocation according to the prediction results and monitoring the actual operation situation in real time. According to the predicted business peak period, the bank rationally arranges resources, including: when the business peak period is predicted to come, increase service windows or customer service staff in advance to reduce customers' waiting time; Dynamically adjust the configuration of system servers based on business forecast results, including increasing system server resources to cope with concurrent requests when business peaks arrive; Comparing the forecast results with the actual business volume helps managers judge the accuracy of the model forecast. If the forecast deviates greatly from the actual situation, the forecast accuracy can be further improved by adjusting the model or optimizing features.

10. A banking business peak period prediction system based on a real-time computing platform, characterized in that: include: The data collection module is used to collect historical transaction data and external factor data from the bank database in real time, and clean and pre-process the collected data; A data processing module is used to extract the pre-processed time, historical window and external factor features, and perform feature coding and engineering optimization processing on the features; The model training and prediction module is used to build a machine learning model, train the model using the encoded time, historical window, and external factor features as input, adjust hyperparameters through cross-validation, evaluate model performance, receive real-time data, extract input features, input real-time data into the trained model for inference, and output prediction results; The result application module is used to dynamically adjust the bank's resource allocation according to the prediction results and reversely optimize the model by real-time monitoring of actual operations.

Images