Financial institution supervision prediction model evaluation system based on cost sensitive learning
Through the financial institution regulatory prediction model evaluation system based on cost-sensitive learning, the problem of difficult model performance under the imbalanced data set is solved, and more accurate and efficient risk supervision of financial institutions is achieved.
Patent Information
- Application Number
- CN202311285444.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-10-07
- Publication Date
- 2025-07-29
AI Technical Summary
When the existing financial institution risk prediction system deals with unbalanced data sets, it is difficult to distinguish the performance of different prediction models, and the existing evaluation standards fail to effectively consider the actual cost differences, resulting in high cost and inaccurate evaluation of model performance.
The regulatory prediction model evaluation system of financial institutions based on cost-sensitive learning is adopted, including a quantitative risk supervision database, demand acceptance unit, numerical extraction unit, data cleaning unit, feature derivative unit, training set generation unit, modeling unit and evaluation unit. The model performance is quantified through the cost-sensitive evaluation method and a regulatory prediction model suitable for financial institutions is constructed.
It improves the accuracy and efficiency of model performance evaluation, reduces labor and time costs, and improves the degree of automation and supervision of financial institutions.
Smart Images

Figure CN120387095A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of financial institution supervision, and more particularly to a financial institution supervision prediction model evaluation system based on cost-sensitive learning. Background Art
[0002] Many real-world classification problems are represented by highly imbalanced datasets, where the number of instances of one class is much smaller than that of the other class. Such problems are called class imbalance learning problems or rare event detection problems. Most traditional classification models tend to provide biased predictions, only focusing on the accuracy of the majority class and ignoring rare events. Many real-world machine learning applications need to build models using highly imbalanced datasets. However, machine learning algorithms assume that misclassification errors are equal, and this assumption is dangerous in imbalanced classification problems. One important reason is that the cost of misclassifying positive samples is higher than that of misclassifying negative samples.
[0003] From a machine learning perspective, the task of detecting enterprise risk prediction is a binary classification problem. Classification models are usually trained and evaluated based on statistical performance metrics, which have two drawbacks. One is that these measures do not take into account the actual business objectives. Enterprise bankruptcy prediction is considered an instance-dependent cost-sensitive classification problem, where the cost due to misclassification varies from instance to instance, and appropriate methods are needed to build the classification model. The other is that when evaluating and comparing the performance of different algorithms, when the statistical performance metrics of different algorithms are all high and perform well in the same scenario, sometimes the statistical performance metrics alone cannot distinguish the algorithms.
[0004] The commonly used evaluation criteria in financial institution risk prediction systems are accuracy and error rate based on the confusion matrix. These two commonly used criteria assign equal weights to the cost of the first type of error and the cost of the second type of error. Nevertheless, the losses caused by bankrupt financial institutions vary greatly in practice. In this case, it is unrealistic to either assign equal weights to the actual costs of the first type of error and the second type of error or simply assume that the actual cost of the first type of error is several times that of the second type of error. Summary of the Invention
[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a financial institution supervision prediction model evaluation system based on cost-sensitive learning to solve the technical problems raised in the background art.
[0006] To achieve the above object, the present invention provides the following technical solutions: A financial institution supervision prediction model evaluation system based on cost-sensitive learning, including a quantitative risk supervision database, a demand acceptance unit, a numerical extraction unit, a data cleaning unit, a feature derivation unit, a training set generation unit, a modeling unit, an evaluation unit, and a callable feature screening module. The demand acceptance unit is used to accept the supervision prediction demand of the target financial institution. The numerical extraction unit is used to extract financial supervision data from the quantitative risk supervision database and intelligently screen features to intelligently construct effective features. The data cleaning unit is used to process the missing values of the features screened in the numerical extraction unit. The feature derivation unit derives new features based on the screened features. The training set generation unit is used to divide the supervision data set into a training set and a test set according to the 20 / 80 rule. The modeling unit is used to apply the training set to input into the classification algorithm toolbox to train and construct a financial institution supervision prediction model. The evaluation unit uses a cost-sensitive evaluation method to quantify the prediction performance of the financial institution supervision prediction model.
[0007] In a preferred embodiment, the quantitative risk supervision database is a database of financial supervision information with supervision attributes. The financial supervision information includes industrial and commercial information, tax information, and the corresponding financial institution types. And the information included in all historical demands in the quantitative risk supervision database constitutes a financial institution information library.
[0008] In a preferred embodiment, when the demand acceptance unit is used to accept the supervision prediction demand of the target financial institution, the supervision prediction request of the target financial institution includes the supervision information of the target financial institution and the type of the financial institution.
[0009] In a preferred embodiment, the numerical extraction unit respectively extracts a numerical data set conforming to numerical features and a character data set conforming to character features from the historical financial institution information library. According to the training set of the supervision prediction model, a group of classification algorithms corresponding to the target training set are selected in a preset model table for prediction and supervision, and this group of financial institution supervision prediction models is determined as the target financial institution supervision prediction model group currently corresponding to the target financial institution. The evaluation unit will select the model with the lowest usage cost from this group.
[0010] In a preferred embodiment, the data cleaning unit removes the instances with missing features, searches for and obtains the outliers in each feature column, and uses the average of all numbers except the outliers in the feature with outliers to replace the outliers. The determination formula for outliers is Q1 - 1.5IQR < x i < Q3 + 1.5IQR, where x i is the feature in the data set, and {x1, x2,..., x i ,..., x n} ∈ X, 1 ≤ i ≤ n, where X is the data set, IQR is the interquartile range of the data set, Q1 is the first quartile, and Q3 is the third quartile.
[0011] In a preferred embodiment, the feature screening module obtains a training set based on the numerical data set collected in the numerical extraction unit. The training set is obtained by using a decision tree to calculate and enumerate the gain of each attribute for each instance, splitting at the point of maximum gain, and weighted summing the gain values of each feature in all decision trees, then dividing by the number of times the feature is used in all decision trees to obtain an average. The feature splitting method of the average gain is used to construct the target financial supervision data set.
[0012] In a preferred embodiment, the formula for calculating and enumerating the gain of each attribute for each instance by the decision tree is In the formula, GL is the sum of the first-order gradients of the loss functions of all samples on the left branch after splitting at a specific node on the decision tree, HL is the sum of the second-order gradients of the loss functions of all samples on the left branch, GR is the sum of the first-order gradients of the loss functions of all samples on the right branch after splitting at a specific node on the decision tree, HR is the sum of the first-order gradients of the loss functions of all samples on the right branch, G = GL + GR, and H = HL + HR.
[0013] In a preferred embodiment, the feature derivation unit uses derived statistical values, arithmetic operations, differences, and logarithms to derive new features. After the new features are derived, the feature screening module is called to further screen based on the new features derived by the feature derivation unit, and together with the previous screening results, they constitute the target supervision data set.
[0014] In a preferred embodiment, the evaluation module uses a cost-sensitive evaluation method, and the method is as follows: Step S1, determine the sample data set; Step S2, define the cost-sensitive matrix; Step S3, define the performance evaluation method; Step S4, evaluate and rank.
[0015] A financial institution supervision prediction model evaluation system based on cost-sensitive learning, and its financial institution supervision prediction model evaluation method based on cost-sensitive learning is as follows:
[0016] Step 1, construct a quantitative risk supervision database based on big data and store it in the quantitative risk supervision data collection module;
[0017] Step 2, the demand acceptance unit accepts the supervision prediction demand for the target financial institution;
[0018] Step 3, the numerical extraction unit extracts a numerical data set conforming to numerical features and a character data set conforming to character features from the historical financial institution information database respectively;
[0019] Step 4: The numerical extraction unit selects a set of classification algorithms corresponding to the target training set from the preset model table according to the training set of the regulatory prediction model for prediction and supervision, and determines this set of financial institution regulatory prediction models as the target financial institution regulatory prediction model currently corresponding to the target institution;
[0020] Step 5: Extract financial supervision data from the quantitative risk supervision database, intelligently screen features, and intelligently construct effective features to obtain multiple groups of financial supervision data groups in the financial supervision information with supervision attributes;
[0021] Step 6: Use the financial supervision element data group as training data to train and obtain an intelligent supervision model;
[0022] Step 7: Collect supervision element data and input it into the intelligent supervision model to obtain a supervision prediction result, and input it into the evaluation module for intelligent evaluation of the prediction result.
[0023] Technical effects and advantages of the present invention:
[0024] 1. The present invention proposes a novel cost-sensitive evaluation method, which effectively solves the problem that it is difficult to discriminate and distinguish the performance of different prediction models in the cost-sensitive problem, thus enriching the perspective of algorithm performance evaluation, effectively reducing the labor cost and time cost of model performance evaluation in the cost-sensitive problem, and ultimately promoting the improvement of the performance of classification algorithms;
[0025] 2. The financial supervision data group of the present invention includes various types of financial supervision element data. Using the financial supervision element data group as training data, a financial institution regulatory prediction model is trained. The financial institution regulatory prediction model is used for intelligent supervision of the risks of financial institutions. Collect supervision element data to obtain a supervision prediction result, and input it into the evaluation module for intelligent evaluation of the prediction result. By intelligent modeling, the automation degree of financial institution risk supervision is improved, and the problems occurring in the process of financial institution supervision are solved pertinently, thereby improving the supervision level of the financial institution industry. BRIEF DESCRIPTION OF THE DRAWINGS
[0026] Figure 1 It is a schematic diagram of the composition of the financial institution regulatory prediction model evaluation system of the present invention.
[0027] Figure 2 It is a schematic diagram of the process of the financial institution regulatory prediction model evaluation method of the present invention. DETAILED DESCRIPTION OF THE INVENTION
[0028] The technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the present invention. In addition, the forms of the various structures described in the following embodiments are merely examples, and a financial institution supervision prediction model evaluation system based on cost-sensitive learning involved in the present invention is not limited to the various structures described in the following embodiments. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.
[0029] Referring to Figure 1 , the present invention provides a financial institution supervision prediction model evaluation system based on cost-sensitive learning, including a quantitative risk supervision database, a demand acceptance unit, a numerical extraction unit, a data cleaning unit, a feature derivation unit, a training set generation unit, a modeling unit, an evaluation unit, and a callable feature screening module. The demand acceptance unit is used to accept the supervision prediction demand of the target financial institution. The numerical extraction unit is used to extract the financial supervision data in the quantitative risk supervision database and intelligently screen features to intelligently construct effective features. The data cleaning unit is used to process the missing values of the features screened in the numerical extraction unit. The feature derivation unit derives new features based on the screened features. The training set generation unit is used to divide the supervision data set into a training set and a test set according to the 20 / 80 rule. The modeling unit is used to apply the training set to input into the classification algorithm toolbox to train and construct a financial institution supervision prediction model. The evaluation unit uses a cost-sensitive evaluation method to quantify the prediction performance of the financial institution supervision prediction model.
[0030] Furthermore, the quantitative risk supervision database is a database of financial supervision information with supervision attributes. The financial supervision information includes industrial and commercial information, tax information, and the corresponding types of financial institutions. And the information included in all historical demands in the quantitative risk supervision database constitutes the financial institution information database. The quantitative risk supervision database includes various information and contains a large amount of data, which increases the accuracy of financial supervision. And the financial institution information database includes all historical demands. Therefore, historical information gradually accumulates, making the system of the present application gradually accumulate data during use, and thus facilitating continuous development.
[0031] Furthermore, when the demand acceptance unit is used to accept the supervision prediction demand of the target financial institution, the supervision prediction request of the target financial institution includes the supervision information and the type of the financial institution, including the supervision information and the type of the target financial institution, which can quickly understand the supervision requirements that the target financial institution needs to carry out, and thus improve the processing speed of the system.
[0032] Further, the numerical value extraction unit extracts a numerical data set conforming to the numerical type feature and a character data set conforming to the character type feature from the historical financial institution information database respectively. According to the training set of the supervision prediction model, a set of classification algorithms corresponding to the target training set is selected in the preset model table for prediction and supervision, and this set of financial institution supervision prediction models is determined as the target financial institution supervision prediction model group corresponding to the target financial institution currently. The evaluation unit selects the model with the lowest usage cost from this group.
[0033] In the embodiment of the present application, the pre-trial model table is as follows:
[0034]
[0035]
[0036] Select the data set according to the relationship between the historical data set and the number of features in the demand acceptance unit. When the number of features / the number of training samples ≥ 10, select weak learners and linear models; when the number of features / the number of training samples ∈ [0.01, 0.1], select non-linear learners; when the number of training samples is much larger than the number of features, select any model. Select a set of classification algorithms corresponding to the target training set in the preset model table for prediction and supervision, and its system response speed is faster.
[0037] Further, the data cleaning unit removes the instances with missing features, and finds and obtains the outliers in each feature column. The average value of all numbers except the outliers in the feature with outliers is used to replace the outliers. The determination formula of the outliers is Q1 - 1.5IQR < x i <Q3 + 1.5IQR, where x i is the feature in the data set, and {x1, x2,..., x i ,..., x n} ∈ X, 1 ≤ i ≤ n, X is the data set, IQR is the interquartile range of the data set, Q1 is the first quartile, Q3 is the third quartile. By removing the instances with missing features and replacing the outliers with the average value, the deviation of the feature itself will not be too large, and thus the final result is ensured to be more consistent with the actual situation.
[0038] Further, the feature screening module obtains the training set based on the numerical data set collected by the numerical value extraction unit. The way to obtain the training set is to calculate and enumerate the gain of each attribute and each instance using the decision tree, split at the maximum gain, and sum the gain values of each feature in all decision trees weighted, and divide by the number of times all decision trees use this feature to obtain the average value. The feature splitting method of the average gain is used to construct the target financial supervision data set. The formula for calculating and enumerating the gain of each attribute and each instance using the decision tree is Where GL is the sum of the first-order gradients of the loss functions of all samples on the left branch after splitting at a specific node on the decision tree, HL is the sum of the second-order gradients of the loss functions of all samples on the left branch, GR is the sum of the first-order gradients of the loss functions of all samples on the right branch after splitting at a specific node on the decision tree, HR is the sum of the first-order gradients of the loss functions of all samples on the right branch, G = GL + GR, and H = HL + HR.
[0039] Further, the feature derivation unit uses derived statistical values, arithmetic operations, differences, and logarithms to derive new features. After the new features are derived, the feature screening module is called to further screen the new features derived by the feature derivation unit and jointly form the target supervision data set with the previous screening results.
[0040] In the embodiment of the present application, the derivation formula of the difference in the feature derivation unit is:
[0041]
[0042] Where k is a certain feature, such as the accounts receivable with a term of 6 months, and x k is the exact value of this feature of a certain sample; k + 1 is the accounts receivable with a term of 9 months, and k - 1 is the accounts receivable with a term of 3 months.
[0043] Further, the evaluation module uses cost-sensitive to evaluate the prediction performance of the financial institution supervision prediction model, and the method is as follows: Step S1, determine the sample data set; Step S2, define the cost-sensitive matrix; Step S3, define the performance evaluation method; Step S4, rank and divide.
[0044] In the embodiment of the present application, the detailed evaluation process of the evaluation module is:
[0045] Step S1, determine the sample data set, and judge whether the misclassification costs of different samples are the same. There are the following three results: Case 1, if the misclassification costs of each sample in a type of error are different, then define the cost-sensitive matrix based on different samples; Case 2, if the misclassification costs between samples are the same, but only the misclassification costs of type I and type II errors are different, then define the cost-sensitive matrix based on the error categories; Case 3, if the misclassification costs of type I errors are the same as those of type II errors, then the statistical performance indicators can be directly used for comparison;
[0046] Step S2, define the cost-sensitive matrix, and define the cost of each prediction result in the cost-sensitive matrix according to the sample data set; in different types of cost-sensitive problems, define the cost matrix based on classes, or define the misclassification cost matrix according to each individual sample;
[0047] The cost matrix is:
[0048] Predict Positive Predict Negative Actual Positive <![CDATA[C(0,0) = C 00 > <![CDATA[C(0,1) = C 01 > Actual Negative <![CDATA[C(1,0) = C 10 > <![CDATA[C(1,1) = C 11 >
[0049] C in the cost matrix 00 is the classification cost for correctly classifying a financial institution, C 01 is the cost of misclassifying a troubled financial institution as a healthy financial institution (type I error cost), C 10 is the cost of misclassifying a healthy financial institution as a troubled financial institution (type II error cost), C 11 is the cost of correctly classifying a troubled financial institution;
[0050] Step S3: Define an evaluation method according to the cost matrix in Step S2. The evaluation principle for Case 1 can be defined as:
[0051] C = C pq (RC i ) + ∑C xy ; (p, q) = (0, 1),
[0052] where C 01 is the misclassification cost of false negative errors, C 10 is the misclassification cost of false positive errors, RC i is the monetary cost required to dispose of Bank i. The sum of the misclassification costs of all samples is defined as the cost-sensitive evaluation method based on samples.
[0053] The evaluation principle for Case 2 can be summarized as
[0054] C = a·C 01 + b·C 10
[0055] (i = 1,..., a; j = 1,..., b)
[0056] where a is the total number of samples of type I errors; b is the total number of samples of type II errors;
[0057] The ranking in Step S4 is obtained by calculating the total cost of the machine learning prediction model results according to the cost-sensitive evaluation method defined in Step S3 and sorting.
[0058] Refer to Figure 2 , a financial institution supervision prediction model evaluation system based on cost-sensitive learning. The financial institution supervision prediction model evaluation method based on cost-sensitive learning is as follows:
[0059] Step 1: Construct a quantitative risk supervision database based on big data and store it in the quantitative risk supervision data acquisition module;
[0060] Step 2: The demand acceptance unit accepts the supervision prediction demand for the target financial institution;
[0061] Step 3: The numerical extraction unit extracts a numerical data set that conforms to the numerical feature type and a character data set that conforms to the character feature type from the historical financial institution information database respectively;
[0062] Step 4: The numerical extraction unit selects a set of classification algorithms corresponding to the target training set from a preset model table according to the training set of the regulatory prediction model for prediction and supervision, and determines this set of financial institution regulatory prediction models as the target financial institution regulatory prediction model corresponding to the target institution currently;
[0063] Step 5: By extracting financial supervision data from the quantitative risk supervision database and intelligently screening features and intelligently constructing effective features, multiple groups of financial supervision data sets in the financial supervision information with supervision attributes are obtained;
[0064] Step 6: Using the financial supervision element data set as training data, a wisdom supervision model is trained;
[0065] Step 7: Collect supervision element data and input it into the wisdom supervision model to obtain a supervision prediction result, and input it into the evaluation module for intelligent evaluation of the prediction result.
[0066] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions according to the embodiments of the present application are generated in whole or in part.
[0067] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.
[0068] In several embodiments provided by this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of the devices or units can be in electrical, mechanical, or other forms.
[0069] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0070] In addition, in each embodiment of this application, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.
[0071] As mentioned above, the above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed by this application can easily think of changes or substitutions, which should all be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.
[0072] Finally: The above is only the preferred embodiment of the present invention and is not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A financial institution supervision prediction model evaluation system based on cost-sensitive learning, characterized in that: It includes a quantitative risk supervision database, a demand acceptance unit, a numerical extraction unit, a data cleaning unit, a feature derivation unit, a training set generation unit, a modeling unit, an evaluation unit, and a callable feature screening module. The demand acceptance unit is used to accept the supervision prediction demands of the target financial institution. The numerical extraction unit is used to extract the financial supervision data in the quantitative risk supervision database, intelligently screen features, and intelligently construct effective features. The data cleaning unit is used to handle the missing values of the features screened in the numerical extraction unit. The feature derivation unit derives new features based on the screened features. The training set generation unit is used to divide the supervision data set into a training set and a test set according to the 20 / 80 rule. The modeling unit is used to apply the training set to input into the classification algorithm toolbox to train and construct a financial institution supervision prediction model. The evaluation unit uses the cost-sensitive evaluation method to quantify the prediction performance of the financial institution supervision prediction model.
2. The evaluation system for the regulatory prediction model of financial institutions based on cost-sensitive learning according to claim 1, wherein: The quantitative risk supervision database is a database of financial supervision information with supervision attributes. The financial supervision information includes industrial and commercial information, tax information, and the corresponding financial institution types. And the information included in all historical demands in the quantitative risk supervision database constitutes the financial institution information database.
3. The evaluation system for the regulatory prediction model of financial institutions based on cost-sensitive learning according to claim 1, wherein: When the demand acceptance unit is used to accept the supervision prediction demands of the target financial institution, the supervision prediction request of the target financial institution includes the supervision information of the target financial institution and the type of the financial institution.
4. The evaluation system for the regulatory prediction model of financial institutions based on cost-sensitive learning according to claim 1, characterized in that: The numerical extraction unit respectively extracts a numerical data set conforming to numerical features and a character data set conforming to character features from the historical financial institution information database. According to the training set of the supervision prediction model, a group of classification algorithms corresponding to the target training set are selected in the preset model table for prediction and supervision, and this group of financial institution supervision prediction model groups corresponding to the current target financial institution are determined. The evaluation unit will select the model with the lowest cost from this group.
5. The evaluation system for a financial institution supervision prediction model based on cost-sensitive learning according to claim 1, characterized in that: The data cleaning unit removes instances with missing features, finds and identifies outliers in each feature column, and replaces the outliers with the average of all numbers except the outliers in the features with outliers. The determination formula for outliers is Q1 - 1.5IQR < x i < Q3 + 1.5IQR, where x i is a feature in the dataset, and {x1, x2,..., x i ,..., x n} ∈ X, 1 ≤ i ≤ n, X is the dataset, IQR is the interquartile range of the dataset, Q1 is the first quartile, and Q3 is the third quartile.
6. The evaluation system for the regulatory prediction model of financial institutions based on cost-sensitive learning according to claim 1, wherein: The feature screening module obtains a training set based on the numerical data set collected in the numerical extraction unit. The way to obtain its training set is to use the decision tree to calculate and enumerate the gain of each attribute for each instance, split at the maximum gain, and sum the gain values of each feature in all decision trees weighted, divide by the number of times this feature is used in all decision trees to get the average value. The feature splitting method of the average gain is used to construct the target financial supervision data set.
7. The evaluation system for the regulatory prediction model of financial institutions based on cost-sensitive learning according to claim 6, characterized in that: The gain calculation formula for each attribute and each instance calculated and enumerated by the decision tree is where GL is the sum of the first-order gradients of the loss functions of all samples on the left branch after splitting at a specific node on the decision tree, HL is the sum of the second-order gradients of the loss functions of all samples on the left branch, GR is the sum of the first-order gradients of the loss functions of all samples on the right branch after splitting at a specific node on the decision tree, HR is the sum of the first-order gradients of the loss functions of all samples on the right branch, G = GL + GR, and H = HL + HR.
8. The evaluation system for the regulatory prediction model of financial institutions based on cost-sensitive learning according to claim 1, wherein: The feature derivation unit uses derivative statistical values, arithmetic operations, differences, and logarithms to derive new features. After the new features are derived, the feature screening module is called again to further screen based on the new features derived by the feature derivation unit and jointly constitute the target supervision data set with the previous screening results.
9. The evaluation system for the regulatory prediction model of financial institutions based on cost-sensitive learning according to claim 1, wherein: The cost-sensitive evaluation method adopted by the evaluation module is as follows: Step S1, determine the sample data set; Step S2, define the cost-sensitive matrix; Step S3, define the performance evaluation method; Step S4, evaluate and rank.
