Business risk prediction system and method combined with multi-source data

Through the business risk prediction system and methods of multi-source data, the problem of limitations in the accuracy and comprehensiveness of evaluation results caused by the single data source is solved, and the accuracy and reliability of risk prediction is improved, providing enterprises with more effective risk identification and Management plan.

CN119047843APending Publication Date: 2024-11-29民信惠弘锋(深圳)投资合伙企业(有限合伙)
View PDF 0 Cites 7 Cited by

Patent Information

Application Number
CN202411313708.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-20
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

Due to the single data source in the prior art, the accuracy and comprehensiveness of the business risk assessment results are limited, which affects the timely identification and effective response of business risks by enterprises.

Method used

Using a business risk prediction system and method combining multi-source data, through data collection, correlation analysis, feature extraction and meta-learner integration, multiple risk prediction sub-models are built and prediction results are integrated to generate more accurate and comprehensive business risk predictions. information.

Benefits of technology

It improves the accuracy and reliability of business risk prediction, provides more effective risk prediction information, and helps enterprises better identify and manage risks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119047843A_ABST
    Figure CN119047843A_ABST
Patent Text Reader

Abstract

The invention provides a business risk prediction system and method combined with multi-source data, and relates to the technical field of risk prediction, and the system comprises a data collection module which carries out the data collection, and generates a multi-source data set; the correlation analysis module is used for performing correlation analysis and generating a data correlation factor; the feature extraction module is used for extracting features and constructing a plurality of risk prediction sub-models; the prediction result output module is used for predicting and outputting a plurality of risk prediction results; and the risk prediction information acquisition module inputs the plurality of risk prediction results into the meta-learner for training, and acquires business risk prediction information. Through the method and the device, the technical problem that in the prior art, due to single data source, accuracy and comprehensiveness of an evaluation result are limited, and timely identification and effective response of an enterprise to a business risk are further influenced can be solved, prediction accuracy and reliability are improved, and more effective risk prediction information is provided for the business.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of risk prediction, and in particular to a business risk prediction system and method combining multi-source data. Background Art

[0002] With the advent of the big data era, the data sources that enterprises can obtain are becoming more and more diverse, including but not limited to market data, user behavior data, financial data, etc. These data can reflect the operating conditions and external environment of the enterprise from multiple perspectives, providing a rich information basis for business risk prediction. In an environment of increasingly fierce market competition, enterprises need to predict business risks in a timely and accurate manner in order to make reasonable decisions and reduce potential losses. Multi-source data can provide a more comprehensive view of risks, helping enterprises better identify and manage risks.

[0003] Currently, existing risk assessment and prediction methods often rely only on a single data source or simple data fusion technology.

[0004] In summary, the existing technology has limited accuracy and comprehensiveness of assessment results due to the single data source, which further affects the company's ability to timely identify and effectively respond to business risks. Summary of the invention

[0005] The purpose of this application is to provide a business risk prediction system and method that combines multi-source data to solve the technical problem that the accuracy and comprehensiveness of the evaluation results are limited due to the single data source in the existing technology, which further affects the company's timely identification and effective response to business risks.

[0006] In view of the above problems, the present application provides a business risk prediction system and method combining multi-source data.

[0007] In the first aspect, the present application also provides a business risk prediction system combining multi-source data, which is used to execute a business risk prediction method combining multi-source data, wherein the system includes: a data acquisition module, the data acquisition module is used to obtain multiple data sources, perform data acquisition based on the multiple data sources, and generate a multi-source data set; an association analysis module, the association analysis module is used to perform association analysis based on the multi-source data set, generate data association factors, transmit the multi-source data set to a data fusion channel according to the data association factors, and determine the multi-source fusion data set; a feature extraction module, the feature extraction module is used to extract features based on the multi-source fusion data set, and construct multiple risk prediction sub-models based on the multi-source fusion features; a prediction result output module, the prediction result output module is used to synchronize the target business data to the multiple risk prediction sub-models for prediction, and output multiple risk prediction results; a risk prediction information acquisition module, the risk prediction information acquisition module is used to integrate the multiple risk prediction results into a meta-learner for training, and obtain business risk prediction information.

[0008] In the second aspect, the present application provides a business risk prediction method combining multi-source data, and the method is implemented by a business risk prediction system combining multi-source data, wherein the method includes: obtaining multiple data sources, collecting data based on the multiple data sources, and generating a multi-source data set; performing correlation analysis based on the multi-source data set to generate data correlation factors, transmitting the multi-source data set to a data fusion channel according to the data correlation factors, and determining the multi-source fusion data set; performing feature extraction based on the multi-source fusion data set, and constructing multiple risk prediction sub-models based on the multi-source fusion features; synchronizing the target business data to the multiple risk prediction sub-models for prediction, and outputting multiple risk prediction results; integrating the multiple risk prediction results and inputting them into a meta-learner for training to obtain business risk prediction information.

[0009] One or more technical solutions provided in this application have at least the following technical effects or advantages: Through the data acquisition module, the data acquisition module is used to obtain multiple data sources, perform data acquisition based on the multiple data sources, and generate a multi-source data set; the association analysis module is used to perform association analysis based on the multi-source data set, generate data association factors, and transmit the multi-source data set to the data fusion channel according to the data association factors to determine the multi-source fusion data set; the feature extraction module is used to extract features based on the multi-source fusion data set, and construct multiple risk prediction sub-models based on the multi-source fusion features; the prediction result output module is used to synchronize the target business data to the multiple risk prediction sub-models for prediction, and output multiple risk prediction results; the risk prediction information acquisition module is used to integrate the multiple risk prediction results into the meta-learner for training, and obtain business risk prediction information, which effectively solves the technical problem that the accuracy and comprehensiveness of the evaluation results are limited due to the single data source in the prior art, further affecting the timely identification and effective response of enterprises to business risks, improves the accuracy and reliability of the prediction, and thus provides more effective risk prediction information for the business.

[0010] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented according to the contents of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are specifically cited below. It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easy to understand through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the present application or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0012] Figure 1 A flowchart of a business risk prediction method combining multi-source data for this application; Figure 2 This is a schematic diagram of the structure of the business risk prediction system that combines multi-source data in this application.

[0013] Description of reference numerals: Data collection module 11, association analysis module 12, feature extraction module 13, prediction result output module 14, risk prediction information acquisition module 15. DETAILED DESCRIPTION

[0014] This application provides a business risk prediction system and method that combines multi-source data, thereby solving the technical problem in the prior art that the accuracy and comprehensiveness of the assessment results are limited due to a single data source, further affecting the company's ability to promptly identify and effectively respond to business risks, thereby improving the accuracy and reliability of the prediction and providing more effective risk prediction information for the business.

[0015] Below, the technical solutions in the present application will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments of the present application. It should be understood that the present application is not limited to the example embodiments described herein. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application. It should also be noted that, for the convenience of description, only the parts related to the present application are shown in the accompanying drawings, rather than all of them. Example

[0016] Please refer to the attached Figure 1 The present application provides a business risk prediction method combining multi-source data, wherein the method is applied to a business risk prediction system combining multi-source data, and the method specifically comprises the following steps: S1: Acquire multiple data sources, collect data based on the multiple data sources, and generate a multi-source data set.

[0017] Specifically, according to business needs and forecasting goals, determine the types of data sources that need to be collected. This includes internal enterprise databases, public market data, social media information, industry reports, etc. Use ETL tools to automatically extract data from various data sources. For unstructured data, such as text, images, etc., specific technologies such as crawlers and OCR recognition can be used for data collection. Integrate the data collected from different data sources to ensure that the data format and encoding are unified. Internal enterprise data includes sales data, financial data, inventory data, etc. External market data includes stock prices, industry trends, economic policies, etc. Configure ETL tasks to define data source connections, data extraction rules, and data cleaning and conversion logic. Execute ETL tasks to extract data from various data sources to the middle layer for cleaning and conversion. Unify the format and code conversion of data obtained from different data sources, and fill or process missing values ​​and outliers.

[0018] S2: performing association analysis based on the multi-source data set to generate a data association factor, transmitting the multi-source data set to a data fusion channel according to the data association factor, and determining a multi-source fusion data set.

[0019] Specifically, association rule mining techniques, such as Apriori algorithm or FP-Growth algorithm, are used to discover frequent item sets and association rules in the data set. Key association factors are extracted from multi-source data sets through factor analysis techniques, such as principal component analysis or maximum likelihood estimation. These factors can represent the main characteristics and trends of the data set. The extracted association factors are rotated, such as varimax rotation, so that the rotated factors are easier to understand and name, which is helpful for subsequent data fusion and analysis. The generated data association factors are used as the transmission medium to transmit the multi-source data set to the data fusion channel. It can ensure that only highly associated and important data are taken into account during the fusion process. During the transmission process, the data from different data sources are aligned and matched to ensure their correspondence in time, space and attributes. This is achieved through time-based alignment, attribute-based matching and other technologies. In the data fusion channel, the associated data are fused using specific data fusion algorithms, such as weighted average, rule-based fusion or machine learning. This can eliminate data redundancy and inconsistency and improve data accuracy and integrity.

[0020] S3: Extract features based on the multi-source fusion data set, and construct multiple risk prediction sub-models based on the multi-source fusion features.

[0021] Specifically, the multi-source fusion data set is cleaned to handle missing values, outliers, and duplicate values ​​to ensure data quality. The data is normalized or standardized to eliminate the dimensional differences between different features. Statistical tests such as analysis of variance, chi-square test, etc. or machine learning algorithms such as random forest feature importance assessment are used to identify features that have a significant impact on risk prediction. Feature engineering techniques such as polynomial features, interactive features, etc. are applied to create new features. Dimensionality reduction techniques such as PCA, t-SNE, etc. are used to reduce the dimension of features while retaining the main change patterns in the data. The multi-source fusion data set is divided into training set, validation set, and test set. The training set is used to train the model, the validation set is used to adjust the model parameters and select the best model, and the test set is used to evaluate the final performance of the model. According to the nature of the problem and the characteristics of the data, a suitable machine learning algorithm is selected to build a risk prediction sub-model, such as logistic regression, support vector machine, decision tree, random forest, gradient boosting machine, etc. The selected algorithm is trained using the training set, and the model performance is optimized by adjusting the model's hyperparameters. Evaluate the performance of the model on the validation set and select the best performing model as the final risk prediction sub-model. Evaluate the performance of the selected model on the test set and calculate indicators such as accuracy, recall, and F1 score. Construct multiple different risk prediction sub-models.

[0022] S4: Synchronize the target business data to the multiple risk prediction sub-models for prediction, and output multiple risk prediction results.

[0023] Specifically, collect the target business data that needs to be risk predicted. These data have the same characteristics and format as the data set used to train the risk prediction sub-model. Perform necessary preprocessing on the target business data, including data cleaning, missing value processing, outlier detection and processing, etc., to ensure that the data quality meets the requirements of the model input. If the format of the target business data is inconsistent with the format required for the risk prediction sub-model input, format conversion is required to ensure that the data can be correctly input into the model. Establish a data synchronization mechanism to synchronize the preprocessed target business data to each risk prediction sub-model. This is achieved through data pipelines, API calls or other data transmission methods. The synchronized target business data is used as input and passed to each risk prediction sub-model respectively. Each risk prediction sub-model predicts the input target business data based on its own algorithm and parameter settings, and outputs the corresponding risk prediction results.

[0024] S5: Integrate the multiple risk prediction results and input them into a meta-learner for training to obtain business risk prediction information.

[0025] Specifically, first, the prediction results output by multiple risk prediction sub-models are integrated into a dataset, and the prediction results of each sub-model can be used as a feature. The dataset contains real business risk labels, which will be used for supervised learning and training the meta-learner. Select a machine learning algorithm as the meta-learner. Including logistic regression, support vector machine, neural network, etc., initialize the parameters and structure of the meta-learner. The integrated dataset is further divided into a training set and a test set. Use the training set to train the meta-learner. In this process, the meta-learner will learn how to make more accurate risk predictions based on the prediction results of multiple sub-models. Tune the hyperparameters of the meta-learner through the validation set to improve the performance of the model. Evaluate the performance of the meta-learner on the test set. This is done by calculating indicators such as accuracy, recall, and F1 score. Compare the performance of the meta-learner with the performance of the sub-models to verify whether the integration method improves the accuracy of the prediction. Once the meta-learner is trained and its performance is verified, it can be used to predict the risks of new business data. Convert the prediction results of the meta-learner into a format that can be understood by the business, such as risk level or specific risk assessment report.

[0026] Furthermore, step S2 of the present application also includes: The target business data traverses the multiple data sources to perform correlation analysis to obtain a list of related data.

[0027] Based on the related data list, association analysis is performed on the multi-source data sets to generate data association factors.

[0028] A data fusion channel is configured, data alignment is performed on the multi-source data sets based on the data association factor, and data fusion is performed on the multi-source data sets according to the data alignment result through the data fusion channel to generate the multi-source fused data set.

[0029] Specifically, all available data sources are traversed, including databases, APIs, files, data streams, etc. In each data source, statistical methods such as Pearson correlation coefficient, Spearman rank correlation coefficient, etc. or machine learning algorithms are used to analyze the correlation between the target business data and the data in the data source. According to the results of the correlation analysis, the data items that are highly correlated with the target business data are listed to form a list of related data. According to the list of related data, the corresponding data is extracted from each data source to form a multi-source data set. Data mining techniques such as Apriori algorithm, FP-Growth algorithm, etc. or statistical analysis methods such as chi-square test, correlation analysis, etc. are used to identify the association between data items in the multi-source data set. According to the results of the association analysis, factors that can reflect the association between data items are extracted, and these factors will be used for subsequent data fusion. A data fusion channel is established, which can process the input of multi-source data sets and perform data fusion according to preset rules. Before data fusion, the data from different data sources are aligned. Including timestamp alignment, data format conversion, missing value filling and other operations to ensure that the data from different data sources can match in structure and content. In the process of data alignment, the generated data association factors are used to guide the matching and alignment of data to ensure that highly associated data can be accurately matched. According to the characteristics of the data and the purpose of fusion, select the appropriate data fusion algorithm, such as weighted average, voting, Bayesian reasoning, etc. Through the data fusion channel, the aligned multi-source data sets are fused. After data fusion, a multi-source fusion data set with unified structure and rich content is generated, which contains information from multiple data sources.

[0030] Furthermore, step S3 of the present application also includes: The data distribution information is obtained by calculation based on the multi-source fusion data set.

[0031] Data mining is performed on the multi-source fusion data set according to the data distribution information to obtain an initial feature set.

[0032] The initial features are randomly extracted based on the random forest algorithm to generate multiple random sample features.

[0033] An evaluation is performed based on the multiple random samples, and initial features are screened based on the evaluation results to generate the multi-source fusion features.

[0034] Specifically, statistical analysis is performed on each feature in the multi-source fusion data set, including calculation of the feature's mean, median, mode, standard deviation, skewness, kurtosis and other statistics. Data is visualized using histograms and box plots. Possible outliers or outliers are identified and processed based on data distribution information. New features or feature combinations are constructed based on data distribution information and business knowledge. Nonlinear features are transformed, such as logarithmic transformation, Box-Cox transformation, etc. The original features and the constructed and transformed new features are merged to form an initial feature set. The characteristics of the random forest algorithm are used, that is, each time a decision tree is constructed, a part of the features are randomly selected for splitting. The model is trained using the random forest algorithm, and the importance score of each feature is obtained. Based on the feature subset used by each tree in the random forest, multiple random sample feature combinations can be generated. Each random sample feature combination is evaluated using evaluation criteria such as accuracy, AUC, etc. Based on the evaluation results, the best performing feature combination is selected, or the feature is screened based on its importance score. After screening, the retained features are the multi-source fusion features, which will be used for subsequent model construction and prediction.

[0035] Furthermore, step S4 of the present application also includes: A feature matrix is ​​constructed based on the multi-source fusion features.

[0036] A feature dimension reduction matrix is ​​generated by reducing the dimension of the feature matrix.

[0037] Random extraction is performed in the feature dimension reduction matrix to obtain a first feature point, a second feature point, ... an Nth feature point, where N is the total number of feature points in the feature dimension reduction matrix.

[0038] Based on the feature dimension reduction matrix, vertical column data and horizontal column data of the first feature point, the second feature point, ... the Nth feature point are obtained.

[0039] A first training data set and a first supervision data set are obtained according to the vertical column data and the horizontal column data of the first feature point, and a first risk prediction sub-model is constructed.

[0040] A second training data set and a second supervision data set are obtained based on the vertical and horizontal data of the second feature point, and a second risk prediction sub-model is constructed. The process is iterated until an Nth training data set and an Nth supervision data set are obtained based on the vertical and horizontal data of the Nth feature point, and an Nth risk prediction sub-model is constructed.

[0041] The multiple risk prediction sub-models are obtained according to the first risk prediction sub-model, the second risk prediction sub-model...the Nth risk prediction sub-model.

[0042] Specifically, first, the extracted multi-source fusion features are sorted and standardized to ensure that each feature has the same scale or range. All multi-source fusion features are combined into a two-dimensional feature matrix, in which each column represents a feature and each row represents a data sample. A dimensionality reduction method is selected, such as principal component analysis, linear discriminant analysis, or t-distribution neighborhood embedding algorithm. The selected dimensionality reduction method is applied to the original feature matrix to obtain a feature dimension reduction matrix with a lower dimension. In the reduced dimension feature matrix, N feature points are randomly selected without replacement. These feature points will be used to construct different risk prediction sub-models. N is the total number of feature points in the feature dimension reduction matrix, or the number selected according to actual needs and resource constraints. For each randomly selected feature point, its corresponding column data, i.e., eigenvalue and column data, i.e., sample identification or label, are extracted. For each feature point, a training data set and a supervision data set are divided according to its column data and column data. The training data set is used to train the model, while the supervision data set is used to verify the accuracy of the model. The corresponding risk prediction sub-model is constructed using the training data set of each feature point and trained. These sub-models can be based on machine learning algorithm models, such as logistic regression, support vector machine, neural network, etc. Repeat the above process until a risk prediction sub-model is built for each feature point. After completing the model construction of all feature points, a set of multiple risk prediction sub-models will be obtained.

[0043] Furthermore, the present application also includes: A matrix point similarity probability distribution function is constructed, and the first feature point of the feature matrix is ​​subjected to dimensionality reduction processing to generate a divergence of the dimensionality reduction similarity of the first feature point.

[0044] A first dimensionality reduction result is generated according to the divergence of the dimensionality reduction similarity of the first feature points.

[0045] The first dimension reduction result is added to the feature dimension reduction matrix.

[0046] Specifically, for each feature point position in the feature matrix, that is, each row or column, calculate its similarity with all other feature point positions, and store these similarity values in a similarity matrix. Convert the values in the similarity matrix into a probability distribution. This is achieved by normalizing the similarity values so that the sum of the similarities of each feature point position is 1, thus forming a probability distribution. Select a dimensionality reduction method, such as PCA, t-SNE, UMAP, etc. Use the first feature point position and its similarity information with other positions as input, and apply the selected dimensionality reduction method for dimensionality reduction. In the reduced-dimensional space, recalculate the similarity between the first feature point position and other positions. The divergence, also known as the difference degree or dispersion degree, can be calculated by comparing the similarities before and after dimensionality reduction. For example, the difference between the similarity matrices before and after dimensionality reduction can be calculated as a measure of divergence. Obtain the coordinates of the first feature point position in the reduced-dimensional space from the dimensionality reduction algorithm. Add this reduced-dimensional coordinate to the corresponding position in the feature dimensionality reduction matrix. If the same processing needs to be performed on other feature point positions, the above steps can be repeated until all required feature point positions are processed and added to the feature dimensionality reduction matrix.

[0047] Furthermore, this application also includes: The matrix point similarity probability distribution function is as follows: ; Wherein, is a measure index representing the proximity between the feature point position of the feature matrix and the reduced-dimensional feature point position of the feature matrix. Let tend to 0 as a constraint condition for dimensionality reduction processing, is the divergence of the similarity between the feature point position of the feature matrix and the reduced-dimensional feature point position of the feature matrix, is the similarity probability based on the reduced-dimensional feature point position of the feature matrix regarding the dot matrix and the dot matrix , is the similarity probability based on the feature point position of the feature matrix regarding the dot matrix and the dot matrix . (i, j) is any coordinate point in the feature matrix, the value range of i is 0 < i < n, the value range of j is 0 < j < n, and n is the total number of point positions in the feature matrix.

[0048] Specifically, The measure index, which represents the measure index of the proximity between the feature point position of the feature matrix and the reduced-dimensional feature point position of the feature matrix. Based on similarity or distance metrics, such as cosine similarity, Euclidean distance, etc. Divergence is used to measure the change in similarity between feature points before and after dimensionality reduction. An ideal dimensionality reduction method should try to maintain the similarity relationship between the original data points, that is, the divergence should be as small as possible. The similarity probability reflects the possibility of observing a specific lattice configuration given the similarity between feature points.

[0049] Furthermore, step S5 of the present application also includes: The meta-learner is constructed based on a logistic regression algorithm.

[0050] The multiple risk prediction results are used as input data, the multiple risk prediction results are classified by the meta-learner, and multiple risk occurrence probabilities are output.

[0051] Risk levels are divided based on target business data to generate risk impact levels.

[0052] The multiple risk prediction results are sorted in descending order based on the risk impact level and the multiple risk occurrence probabilities to generate a risk prediction table.

[0053] The business risk prediction information is determined according to the risk prediction table.

[0054] Specifically, prepare a data set for training the logistic regression model. This data set should contain multiple risk prediction results and the corresponding real risk occurrences. Select features related to risk prediction as inputs to the logistic regression model. Use the logistic regression algorithm to train the model. Logistic regression is a generalized linear model. Evaluate the performance of the model through cross-validation, accuracy, recall, and other indicators. Arrange the risk prediction results obtained by each risk prediction sub-model into a format suitable for input to the meta-learner. Ensure that all risk prediction results are on the same scale so that the meta-learner can correctly classify them. Input the organized risk prediction results into the meta-learner to obtain the risk occurrence probability corresponding to each result. The meta-learner outputs the probability that each risk prediction result belongs to each risk category. Determine the risk level classification criteria based on the specific data and needs of the target business. According to the classification criteria, classify the business data into different risk levels, such as low risk, medium risk, and high risk. Combine the risk occurrence probability output by the meta-learner with the risk level of the business data. Arrange in descending order according to the comprehensive evaluation results of risk occurrence probability and risk level, so as to give priority to high-risk and high-probability events. The results in descending order are sorted into a risk prediction table, which should include information such as risk points, risk levels, and risk probability. Based on the risk prediction table, it is possible to determine which business links or events have higher risks, and thus formulate corresponding risk response strategies. Example

[0055] Based on the business risk prediction method combining multi-source data in the above-mentioned embodiment, the present application also provides a business risk prediction system combining multi-source data, see the attached Figure 2 , the system comprising: The data collection module 11 is used to obtain multiple data sources, collect data based on the multiple data sources, and generate a multi-source data set.

[0056] The association analysis module 12 is used to perform association analysis based on the multi-source data set, generate data association factors, transmit the multi-source data set to a data fusion channel according to the data association factors, and determine a multi-source fusion data set.

[0057] The feature extraction module 13 is used to extract features according to the multi-source fusion data set and construct multiple risk prediction sub-models according to the multi-source fusion features.

[0058] The prediction result output module 14 is used to synchronize the target business data to the multiple risk prediction sub-models for prediction and output multiple risk prediction results.

[0059] The risk prediction information acquisition module 15 is used to integrate the multiple risk prediction results and input them into the meta-learner for training to obtain business risk prediction information.

[0060] Furthermore, the association analysis module 12 in the system further includes a data fusion module, which is used to: The target business data traverses the multiple data sources to perform correlation analysis to obtain a list of related data.

[0061] Based on the related data list, association analysis is performed on the multi-source data sets to generate data association factors.

[0062] A data fusion channel is configured, data alignment is performed on the multi-source data sets based on the data association factor, and data fusion is performed on the multi-source data sets according to the data alignment result through the data fusion channel to generate the multi-source fused data set.

[0063] Furthermore, the feature extraction module 13 in the system further includes a sample evaluation module, which is used to: The data distribution information is calculated and acquired based on the multi-source fusion data set.

[0064] Data mining is performed on the multi-source fusion data set according to the data distribution information to obtain an initial feature set.

[0065] The initial features are randomly extracted based on a random forest algorithm to generate multiple random sample features.

[0066] An evaluation is performed based on the multiple random samples, and initial features are screened based on the evaluation results to generate the multi-source fusion features.

[0067] Furthermore, the prediction result output module 14 in the system further includes a prediction sub-model construction module, and the prediction sub-model construction module is used to: A feature matrix is ​​constructed based on the multi-source fusion features.

[0068] A feature dimension reduction matrix is ​​generated by reducing the dimension of the feature matrix.

[0069] Random extraction is performed in the feature dimension reduction matrix to obtain a first feature point, a second feature point, ... an Nth feature point, where N is the total number of feature points in the feature dimension reduction matrix.

[0070] Based on the feature dimension reduction matrix, vertical column data and horizontal column data of the first feature point, the second feature point, ... the Nth feature point are obtained.

[0071] A first training data set and a first supervision data set are obtained according to the vertical column data and the horizontal column data of the first feature point, and a first risk prediction sub-model is constructed.

[0072] A second training data set and a second supervision data set are obtained based on the vertical and horizontal data of the second feature point, and a second risk prediction sub-model is constructed. The process is iterated until an Nth training data set and an Nth supervision data set are obtained based on the vertical and horizontal data of the Nth feature point, and an Nth risk prediction sub-model is constructed.

[0073] The multiple risk prediction sub-models are obtained according to the first risk prediction sub-model, the second risk prediction sub-model...the Nth risk prediction sub-model.

[0074] Furthermore, the prediction sub-model construction module in the system is also used for: A matrix point similarity probability distribution function is constructed, and the first feature point of the feature matrix is ​​subjected to dimensionality reduction processing to generate a divergence of the dimensionality reduction similarity of the first feature point.

[0075] A first dimensionality reduction result is generated according to the divergence of the dimensionality reduction similarity of the first feature points.

[0076] The first dimension reduction result is added to the feature dimension reduction matrix.

[0077] Furthermore, the prediction sub-model construction module in the system is also used for: Construct a matrix point similarity probability distribution function, and the matrix point similarity probability distribution function is as follows: ; Among them, is a measure index representing the proximity degree between the feature points of the feature matrix and the dimensionality-reduced feature points of the feature matrix. Let tend to 0 as a constraint condition for dimensionality reduction processing, is the divergence of the similarity between the feature points of the feature matrix and the dimensionality-reduced feature points of the feature matrix, is the similarity probability based on the dimensionality-reduced feature points of the feature matrix regarding the dot matrix and the dot matrix , is the similarity probability based on the feature points of the feature matrix regarding the dot matrix and the dot matrix . (i, j) is any coordinate point in the feature matrix, the value range of i is 0 < i < n, the value range of j is 0 < j < n, and n is the total number of points in the feature matrix.

[0078] Furthermore, the risk prediction information acquisition module 15 in the system further includes a risk prediction table generation module, and the risk prediction table generation module is used for: Construct the meta-learner based on the logistic regression algorithm.

[0079] Take the multiple risk prediction results as input data, classify the multiple risk prediction results through the meta-learner, and output multiple risk occurrence probabilities.

[0080] Conduct risk level division based on the target business data to generate a risk impact level.

[0081] Based on the risk impact level and in combination with the multiple risk occurrence probabilities, perform a descending order processing on the multiple risk prediction results to generate a risk prediction table.

[0082] Determine the business risk prediction information according to the risk prediction table.

[0083] In this specification, each embodiment is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The foregoing Figure 1The business risk prediction method and specific examples combining multi-source data in the first embodiment are also applicable to the business risk prediction system combining multi-source data in the present embodiment. Through the above detailed description of the business risk prediction method combining multi-source data, those skilled in the art can clearly understand the business risk prediction system combining multi-source data in the present embodiment, so for the sake of brevity of the specification, it will not be described in detail here. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0084] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

[0085] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the present application and its equivalent technology, the present application is also intended to include these modifications and variations.

Claims

1. A business risk prediction system combining multi-source data, characterized in that: The system comprises: A data collection module, the data collection module is used to obtain multiple data sources, collect data based on the multiple data sources, and generate a multi-source data set; An association analysis module, the association analysis module is used to perform association analysis based on the multi-source data set, generate a data association factor, transmit the multi-source data set to a data fusion channel according to the data association factor, and determine a multi-source fusion data set; A feature extraction module, wherein the feature extraction module is used to extract features according to the multi-source fusion data set and construct multiple risk prediction sub-models according to the multi-source fusion features; A prediction result output module, which is used to synchronize the target business data to the multiple risk prediction sub-models for prediction and output multiple risk prediction results; A risk prediction information acquisition module is used to integrate the multiple risk prediction results and input them into a meta-learner for training to obtain business risk prediction information.

2. The system according to claim 1, characterized in that The association analysis module also includes a data fusion module, which is used to: The target business data traverses the multiple data sources to perform correlation analysis and obtain a list of related data; Performing association analysis on the multi-source data set based on the related data list to generate a data association factor; A data fusion channel is configured, data alignment is performed on the multi-source data sets based on the data association factor, and data fusion is performed on the multi-source data sets according to the data alignment result through the data fusion channel to generate the multi-source fused data set.

3. The system according to claim 1, characterized in that The feature extraction module also includes a sample evaluation module, which is used to: Calculate and obtain data distribution information based on the multi-source fusion data set; Performing data mining on the multi-source fusion data set according to the data distribution information to obtain an initial feature set; Randomly extract the initial features based on a random forest algorithm to generate multiple random sample features; An evaluation is performed based on the multiple random samples, and initial features are screened based on the evaluation results to generate the multi-source fusion features.

4. The system according to claim 1, characterized in that The prediction result output module also includes a prediction sub-model construction module, which is used to: Constructing a feature matrix based on the multi-source fusion features; Generate a feature dimension reduction matrix by reducing the dimension of the feature matrix; Performing random extraction in the feature dimension reduction matrix to obtain a first feature point, a second feature point, ... an Nth feature point, where N is the total number of feature points in the feature dimension reduction matrix; Based on the feature dimension reduction matrix, obtain the vertical column data and the horizontal column data of the first feature point, the second feature point ... the Nth feature point; Acquire a first training data set and a first supervision data set according to the column data and the horizontal column data of the first feature point, and construct a first risk prediction sub-model; A second training data set and a second supervision data set are obtained according to the column data and the horizontal data of the second feature point, and a second risk prediction sub-model is constructed, and iteration is performed until an Nth training data set and an Nth supervision data set are obtained according to the column data and the horizontal data of the Nth feature point, and an Nth risk prediction sub-model is constructed; The multiple risk prediction sub-models are obtained according to the first risk prediction sub-model, the second risk prediction sub-model...the Nth risk prediction sub-model.

5. The system according to claim 4, characterized in that The forecast sub-model building block is also used to: Constructing a matrix point similarity probability distribution function, performing dimensionality reduction processing on the first feature point of the feature matrix, and generating a divergence of the dimensionality reduction similarity of the first feature point; Generate a first dimensionality reduction result according to the divergence of the dimensionality reduction similarity of the first feature point; The first dimension reduction result is added to the feature dimension reduction matrix.

6. The system according to claim 5, characterized in that The forecast sub-model building block is also used to: Construct a matrix point similarity probability distribution function, the matrix point similarity probability distribution function is as follows: ; Among them, is a measurement index for characterizing the proximity between the feature points of the feature matrix and the dimensionality-reduced feature points of the feature matrix. Let tend to 0 as a constraint condition for dimensionality reduction processing. is the divergence of the similarity between the feature points of the feature matrix and the dimensionality-reduced feature points of the feature matrix. is the similarity probability based on the dimensionality-reduced feature points of the feature matrix with respect to the dot matrix and the dot matrix . is the similarity probability based on the feature points of the feature matrix with respect to the dot matrix and the dot matrix . (i, j) is any coordinate point in the feature matrix, where the value range of i is 0 < i < n, the value range of j is 0 < j < n, and n is the total number of feature points in the feature matrix.

7. The system according to claim 1, characterized in that The risk prediction information acquisition module also includes a risk prediction table generation module, which is used to: Constructing the meta-learner based on a logistic regression algorithm; Taking the multiple risk prediction results as input data, classifying the multiple risk prediction results through the meta-learner, and outputting multiple risk occurrence probabilities; Based on the target business data, risk levels are divided and risk impact levels are generated; Based on the risk impact level and the multiple risk occurrence probabilities, the multiple risk prediction results are processed in descending order to generate a risk prediction table; The business risk prediction information is determined according to the risk prediction table.

8. A business risk prediction method combining multi-source data, characterized in that: The method is performed by the system according to any one of claims 1 to 7, and the method comprises: Acquire multiple data sources, collect data based on the multiple data sources, and generate a multi-source data set; Performing association analysis based on the multi-source data set to generate a data association factor, transmitting the multi-source data set to a data fusion channel according to the data association factor, and determining a multi-source fusion data set; Extracting features based on the multi-source fusion data set, and constructing multiple risk prediction sub-models based on the multi-source fusion features; Synchronizing the target business data to the multiple risk prediction sub-models for prediction, and outputting multiple risk prediction results; The multiple risk prediction results are integrated and input into a meta-learner for training to obtain business risk prediction information.

Citation Information

Cited By

  • Dangerous cargo risk early warning method and system and computer program

    CN119359057A

  • Traffic risk dynamic evaluation model construction system

    CN119516791A

  • Telecommunication data integration analysis method and system

    CN119697054A

  • Risk early warning method, system and equipment based on adaptive algorithm selection and medium

    CN119721723A

  • Industrial gathering area pollution risk management method and system based on multi-source data

    CN119990769A