Polar coordinate-based case data set imbalance processing method

By improving the SMOTE algorithm under polar coordinates and generating new minority samples in the sector-shaped area, the blindness and limitations of the existing SMOTE algorithm in medical data processing are solved, and the generalization ability and diagnostic accuracy of the model are improved.

CN119066408APending Publication Date: 2024-12-03ZHONGBEI UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411148375.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-21
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

The existing SMOTE algorithms have blindness and limitations when processing imbalanced data in the medical field, and cannot effectively generate new samples that truly reflect the data distribution, resulting in poor generalization capabilities of the model and may lead to missed diagnosis or misdiagnosis.

Method used

Using SMOTE improved algorithm based on polar coordinates, a new minority sample is randomly generated in the sector-shaped area. Through PCA operation and polar coordinate conversion, the number of pathological samples is increased, data imbalance problem is alleviated, noise impact is reduced, and the quality of generated samples is improved.

Benefits of technology

It improves the generalization ability and diagnostic accuracy of the model, reduces the probability of misdiagnosis and misdiagnosis, enhances the robustness of the model, can better process complex medical data, and improves sample category distribution.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119066408A_ABST
    Figure CN119066408A_ABST
Patent Text Reader

Abstract

The invention discloses a case data set imbalance processing method based on polar coordinates. The method comprises the following steps: S110, collecting a case data set; s120, performing PCA operation to generate a principal component analysis matrix; s130, neighbor samples and majority samples of the minority samples are selected, and polar coordinates are generated; s140, performing random sampling in the polar coordinate system to generate a new sample; s150, converting back to a Cartesian coordinate system, and generating a new sample matrix; s160, adding the generated new samples into a minority sample set, and repeatedly sampling; according to the unbalanced case data set processed by the algorithm provided by the invention, the prediction accuracy and the overall balance of the model for minority class samples can be remarkably improved. The generated new sample is beneficial to reducing the deviation of the model, so that the model can better learn and reflect the characteristics of all samples, and all types of samples, especially pathological samples, can be fairly fitted.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of medical case data mining technology, and in particular to a data set with extremely unbalanced positive and negative samples. The data processing technology under polar coordinates can arbitrarily increase minority class samples, thereby achieving the purpose of improving the performance of machine learning algorithms. Background Art

[0002] Imbalanced data refers to the situation where the number of samples of different categories in a data set is quite different, where the samples with a smaller number are called minority samples and the samples with a larger number are called majority samples. The phenomenon of data imbalance also exists in the medical field. The number of healthy samples and pathological samples is often unbalanced. In some diseases, the number of pathological samples is much smaller than that of healthy samples. When training machine learning models, the imbalance of case data may lead to poor generalization ability of the model and the inability to handle the diverse pathological samples encountered in actual clinical practice, resulting in missed diagnosis or misdiagnosis, which may cause patients to not receive the necessary treatment in time, thus delaying the disease or even aggravating the disease. There are two main solutions to deal with this kind of imbalanced data: one is an algorithm-based solution, including adjusting the classification threshold or cost-sensitive learning, which refers to adding the cost weight of the class imbalance in the classification process, so that the classifier will impose a greater penalty on the misclassified minority class. The representative algorithm is adacost; the other is a data-based solution, mainly a resampling method, including undersampling and oversampling and corresponding improvement methods.

[0003] SMOTE (Synthetic Minority Over-sampling Technique) is a commonly used method for processing imbalanced data. The SMOTE algorithm is widely used in important fields such as medical diagnosis, network intrusion detection, and financial risk assessment. Its core idea is to balance the data set by generating new samples between minority class samples. Specifically, the SMOTE algorithm generates new samples by interpolating minority class samples in the feature space, thereby increasing the number of minority classes. However, since the SMOTE algorithm can only perform linear interpolation on the line connecting the sample and the nearest neighbor point, it is blind and the sample points generated by the minority class samples are limited. At the same time, the linear interpolation method may generate unrealistic samples between minority class samples, especially near the boundary. These samples may not represent the actual data distribution, thereby introducing noise. When there is an intra-class imbalance in the minority class samples, simple linear interpolation to generate samples will aggravate the intra-class imbalance of the samples. The SMOTE algorithm does not consider the boundary between the majority class and the minority class. If the line connecting the minority class sample and the nearest neighbor point is exactly on the boundary, then it is also easy to cause the problem of distribution marginalization.

[0004] The existing SMOTE algorithm and its improved method can only generate new sample points blindly on the line connecting the sample point and the neighboring points. The limitations of the generated positions may cause the generated samples to fail to reflect the distribution of real data well. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a method for processing imbalanced case data sets based on polar coordinates in response to the deficiencies of the prior art.

[0006] The technical solution of the present invention is as follows:

[0007] A method for processing imbalanced case data sets based on polar coordinates. The overall process is as follows: Figure 1 The method includes steps S110-S160, which are as follows:

[0008] S110: Collect case data sets, where case data includes healthy samples and pathological samples. Since the proportion of healthy sample data in the overall data set is relatively large, the healthy sample data is used as the majority class sample data, and the pathological sample data is used as the minority class sample data. The minority class sample data and the majority class sample data constitute the minority sample data set and the majority sample data set respectively;

[0009] S120: Perform PCA operation to generate a principal component analysis matrix. Perform PCA operation without dimensionality reduction on the minority class sample data set to obtain a principal component analysis matrix;

[0010] S130: Select the neighboring samples and majority samples of the minority sample to generate polar coordinates. Randomly select a sample from the minority class sample as the root sample, select its k neighboring samples and a majority sample, and calculate the polar coordinates of the main features;

[0011] S140: Perform random sampling in the polar coordinate system to generate new samples. Randomly select polar coordinate positions in the defined annular range and calculate the feature vector of the generated new sample;

[0012] S150: Convert back to the Cartesian coordinate system to generate a new sample matrix. Convert the polar coordinate position to the Cartesian coordinate position, obtain the eigenvector of the new sample, and form a new principal component analysis matrix;

[0013] S160: Add the generated new samples to the minority sample set and repeat the sampling. Add the newly generated sample matrix to the minority sample set and repeat the above sampling steps until the predetermined number of sampling times is reached.

[0014] The PCA operation process without dimensionality reduction in step S120 is as follows: Figure 2As shown, it includes steps S210-S230. Specifically, S210 randomly selects minority class samples as root samples, S220 performs PCA operation without dimension reduction on matrix X to obtain feature matrix P, and S230 calculates principal component analysis matrix and principal component eigenvector. Each module is described below.

[0015] Let the minority sample set be D, the upsampling factor be n (repeated n times), the number of sample neighbors be k, and the sample distribution be Figure 3 shown.

[0016] S210: Randomly select sample x from the minority class sample i As the root sample for generating new samples, i = 1, 2, ..., |D|. Let the sample x including T features i The eigenvector of Then all samples are represented as matrices Where N is the number of all samples.

[0017] S220: After performing a non-dimensionality reduction PCA operation on X (ie, the length of the eigenvector is still T), a eigenmatrix P is obtained.

[0018] S230: Calculate the principal component analysis matrix X′ and the principal component eigenvector x i ′. Let the principal component eigenvector in X′ be represented by x′ i The formula is as follows:

[0019] X′=XP

[0020] The calculation process of generating polar coordinates in step S130 is as follows: Figure 4 As shown, it includes steps S310-S320. Specifically, S310 selects neighboring samples, and S320 groups and calculates polar coordinates. Each module is described below.

[0021] S310: Select x in D i k nearest neighbor samples, let x ij For sample x i The jth nearest neighbor sample, j = 1, 2, ..., k. And select a majority sample x of the nearest neighbor o ,Right now

[0022] S320: x' i The main features in are grouped in pairs in order, that is, and The two-dimensional Cartesian space points as the d-th group of main features, where d = 1, 2, ..., T-1 (and where and As the two-dimensional Cartesian space point of the main feature of the T group). Similarly, x oThe main eigenvector of is denoted as x' o , and x' o The features in are grouped in pairs in order. o The two-dimensional Cartesian space points grouped in are used as the poles of the polar coordinate system. A ray is drawn in the horizontal positive direction as the polar axis, and the positions of the two-dimensional Cartesian space points of all samples in the polar coordinate system are calculated. Let the main features The polar coordinates are The polar coordinates are calculated as follows:

[0023]

[0024]

[0025] The process of randomly sampling in the polar coordinate system to generate new samples in step S140 is as follows: Figure 5 As shown, it includes steps S410-S420. Specifically, S410 defines a sample generation area, and S420 generates a new sample. Each module is described below.

[0026] S410 and S420: will be based on points and Point The minor arc with endpoints and Point The range enclosed by the minor arc of the endpoint is defined as the sampling range of the dth group of main features, such as Figure 6 As shown. Randomly select polar coordinates in this annular range The calculation expression is as follows:

[0027]

[0028] Where min(·,·) is the function for calculating the minimum value; λ1 and λ2 are random numbers between [0,1].

[0029] The main calculation method of converting back to the Cartesian coordinate system and generating a new sample matrix in step S150 is as follows. The polar coordinates are The Cartesian coordinates are then calculated as follows:

[0030]

[0031] Through calculation, the main features of the dth group in Cartesian coordinates are obtained as follows: and Repeat the above sampling steps d times to get the main feature vector x' after sampling new , forming the principal component analysis matrix X′ new . Further, the sampling matrix X is obtained by inverse PCA calculation. new, the calculation is shown as follows:

[0032] X new =P -1 X′ new

[0033] The sample matrix X new Add to the minority sample set D. Repeat the above steps n times to obtain a new case sample data set D'.

[0034] The improved SMOTE algorithm based on polar coordinates forms new minority class samples in the fan-shaped area and increases the number of pathological samples in the data set, thereby alleviating the problem of data imbalance. Compared with some other SMOTE improved algorithms, such as the Borderline SMOTE algorithm, this algorithm only uses minority class samples on the boundary to synthesize new samples, but the boundary samples may include some noise points. These noise samples may be mistakenly considered as boundary samples, thereby affecting the quality of the generated synthetic samples. The improved SMOTE algorithm based on polar coordinates focuses on all minority class samples and randomly generates new minority class samples in the fan-shaped area between adjacent minority class samples, reducing the impact of noise on the generated samples, improving the quality of the generated samples, and thus improving the category distribution of the samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 : Overall flow chart;

[0036] Figure 2 : PCA operation flow chart without dimensionality reduction;

[0037] Figure 3 : Sample distribution diagram;

[0038] Figure 4 : Generate polar coordinate flow chart;

[0039] Figure 5 : Generate a new sample flow chart;

[0040] Figure 6 : Improved polar coordinate SMOTE algorithm to generate sample diagram;

[0041] Figure 7 : Specific flow chart of calculation of embodiment; DETAILED DESCRIPTION

[0042] The present invention is described in detail below in conjunction with specific embodiments.

[0043] Example 1: Classification method applied to acute ischemic stroke case data

[0044] The specific process in this embodiment is as follows Figure 7 As shown, the specific steps are as follows:

[0045] The acute ischemic stroke case data contains a total of 5110 cases. The 5110 cases of case data are used as the overall sample set of this embodiment, and 6 features are selected as the characteristic attributes of this embodiment, namely "gender", "whether suffering from hypertension", "whether suffering from heart disease", "marital status", "work type" and "average blood sugar level", among which the work type attribute includes private enterprises, government work and freelancers. Among them, the number of majority class samples is 4861, the number of minority class samples is 249, and the imbalance rate IR is 19.52, that is, the number of majority class samples is 19.52 times the number of minority class samples, which is obviously an unbalanced data set.

[0046] First, the data set is preprocessed and the sample x including 6 features is i The eigenvector of Then all samples are represented as matrices Then, perform PCA operation on X without dimensionality reduction to obtain the feature matrix P. After matrix multiplication operation between X matrix and P matrix, the principal component analysis matrix X′ is obtained. Let the principal component eigenvector in X′ be represented as x′ i .

[0047] Select x from the minority class sample D i k nearest neighbor samples, in this embodiment, the nearest neighbor number k is set to 5, let x ij For sample x i The jth nearest neighbor sample, j = 1, 2, ..., 5. And select a majority sample x of the nearest neighbor o ,Right now x' i The main features in are grouped in pairs in order, that is, and The two-dimensional Cartesian space points as the main features of the dth group, where d = 1, 2, ..., 5 ( and As the last group, i.e. the sixth group of main features, the two-dimensional Cartesian space points). Similarly, x o The main eigenvector of is denoted as x′ o , and x′ o The features in are grouped in pairs in order. o The two-dimensional Cartesian space points grouped in the polar coordinate system are used as the poles, and a ray is drawn in the horizontal positive direction as the polar axis to calculate the position coordinates of the two-dimensional Cartesian space points of all samples in the polar coordinate system.

[0048] Then, the point ) and dot The minor arc with endpoints and Point The range enclosed by the minor arc of the endpoint is defined as the sampling range of the dth group of main features, where d = 1, 2, ..., 6. Randomly select polar coordinates in the annular range Then convert the polar coordinates of all samples into Cartesian coordinates. The polar coordinates are

[0049] Through calculation, we can get the main features of the dth group in Cartesian coordinates as follows: and Where d = 1, 2, ..., 6, the main eigenvector x' is obtained after sampling new , forming the principal component analysis matrix X′ new Then X′ new The matrix is ​​further calculated by inverse PCA to obtain the sampling matrix X new Finally, the sample matrix X new Add to the minority sample set D to get D'.

[0050] In this embodiment, the sampling rate is set to 3, and the above steps are repeated 3 times to generate a new sample set.

[0051] Support Vector Machine (SVM) is a supervised learning algorithm widely used in classification and regression analysis. The core idea of ​​SVM is to find a hyperplane to separate data points of different categories as much as possible and maximize the interval between two categories. This hyperplane is called the decision boundary. SVM can perform linear or nonlinear classification and regression tasks, and even outlier detection tasks. It is one of the classic classification models in the field of machine learning. SVM is suitable for the classification of small and medium-sized complex data sets. Moreover, SVM has good robustness and strong generalization ability for unknown data, especially when the amount of data is small, it has better performance than other traditional machine learning algorithms. When the sample data is linearly inseparable, the support vector machine can map the sample data to a linearly separable high-dimensional data space through a kernel function.

[0052] Assume that the equation of the hyperplane is: w·x+b=0, where w is the weight vector, b is the bias term, and x is the input vector.

[0053] The SVM classification problem finds the maximum margin hyperplane by solving the following optimization problem:

[0054]

[0055] sty i (w·x i+b)≥1,i=1,2,…N

[0056] This example uses the support vector machine algorithm to classify the generated new sample set. First, we randomly divide the sample set into a training set and a test set in a ratio of 8:2. To establish the SVM model, we select the radial basis function kernel (RBF) as the kernel function in this example. The RBF kernel function can classify samples in an infinite-dimensional space and is suitable for processing complex classification boundaries. Through the kernel function, the SVM can map the original input data to a higher-dimensional feature space, in which nonlinear problems may become linearly separable. Secondly, the penalty parameter C is used to control the model's tolerance for misclassification. It determines the intensity of the penalty for misclassification. A large penalty parameter will cause the model to try to classify the training data correctly and minimize the error as much as possible, but it will also make the model more complex, while the penalty parameter will make the model simpler and prone to underfitting. The regularization parameter γ is used to control the width of the RBF kernel function. A large γ value will make the model more sensitive to noise, while a small γ value will make the model smoother. This example uses the grid search method to find the optimal regularization parameter and penalty parameter. Finally, the SVM model is trained, the test set is predicted, and the model is evaluated using precision and recall as evaluation indicators.

[0057] The improved SMOTE algorithm based on polar coordinates can alleviate the problem of data imbalance by forming new minority class samples in the fan-shaped area and increasing the number of pathological samples in the data set. Compared with the traditional SMOTE algorithm, which can only generate new sample points by linear interpolation on the line connecting the sample and the nearest neighbor point, it has limitations and cannot truly reflect the complex medical data characteristics. The improved SMOTE algorithm based on polar coordinates takes into account the distribution characteristics of the samples, and the generated sample data is more in line with the actual situation, which improves the generalization ability of the model and the performance of the model on complex medical data. Moreover, through comparison, it is found that the imbalanced case data set processed by the improved SMOTE algorithm based on polar coordinates and then classified by support vector machine can significantly improve the model's recognition ability for minority class pathological samples, improve the classification balance of the model between healthy samples and pathological samples, and thus improve the overall accuracy and sensitivity of diagnosis. Compared with unprocessed imbalanced case data, this method can reduce the probability of missed diagnosis and misdiagnosis, and at the same time, it is more stable when facing new case data, enhances the robustness of the model, and thus helps medical experts diagnose in the decision-making process of pathological diagnosis, which helps to improve the accuracy and timeliness of patient diagnosis.

[0058] Example 2: Regression method applied to acute ischemic stroke case data

[0059] The specific process in this embodiment is as follows Figure 7 As shown, the specific steps are as follows:

[0060] The case data of acute ischemic stroke contains a total of 5110 cases. The 5110 cases of case data are used as the overall sample set of this embodiment, and 10 features are selected as the characteristic attributes of this embodiment, namely "gender", "age", "whether suffering from hypertension", "whether suffering from heart disease", "marital status", "work type", "residence type", "average blood sugar level", "body weight index" and "smoking status", among which the work type attribute includes private enterprises, government work and freelancers; the residence type attribute includes urban and rural areas. Among them, the number of majority class samples is 4861, the number of minority class samples is 249, and the imbalance rate IR is 19.52, that is, the number of majority class samples is 19.52 times the number of minority class samples, which is obviously an unbalanced data set.

[0061] First, the data set is preprocessed to include samples x with 10 features. i The eigenvector of Then all samples are represented as matrices Then, perform PCA operation on X without dimensionality reduction to obtain the feature matrix P. After matrix multiplication operation between X matrix and P matrix, the principal component analysis matrix X′ is obtained. Let the principal component eigenvector in X′ be represented as x′ i .

[0062] Select x from the minority class sample D i k nearest neighbor samples, in this embodiment, the nearest neighbor number k is set to 5, let x ij For sample x i The jth nearest neighbor sample, j = 1, 2, ..., 5. And select a majority sample x of the nearest neighbor o ,Right now x' i The main features in are grouped in pairs in order, that is, and The two-dimensional Cartesian space points as the main features of the dth group, where d = 1, 2, ..., 10 ( and As the last group, i.e. the 10th group of main features, the two-dimensional Cartesian space points). Similarly, x o The main eigenvector of is denoted as x′ o , and x′ o The features in are grouped in pairs in order. o The two-dimensional Cartesian space points grouped in the polar coordinate system are used as the poles, and a ray is drawn in the horizontal positive direction as the polar axis to calculate the position coordinates of the two-dimensional Cartesian space points of all samples in the polar coordinate system.

[0063] Then, the point and Point The minor arc with endpoints and Point The range enclosed by the minor arc of the endpoint is defined as the sampling range of the dth group of main features, where d = 1, 2, ..., 10. Randomly select polar coordinates in the annular range Then convert the polar coordinates of all samples into Cartesian coordinates. The polar coordinates are

[0064] Through calculation, we can get the main features of the dth group in Cartesian coordinates as follows: and Where d = 1, 2, ..., 10, the main eigenvector x' is obtained after sampling new , forming the principal component analysis matrix X′ new Then X′ new The matrix is ​​further calculated by inverse PCA to obtain the sampling matrix X new Finally, the sample matrix X new Add to the minority sample set D to get D'.

[0065] In this embodiment, the sampling rate is set to 5, and the above steps are repeated 5 times to generate a new sample set.

[0066] Multiple Linear Regression is a complex model in regression analysis that considers the impact of multiple input variables on the output variable. Unlike univariate linear regression, multiple linear regression models the system relationship more comprehensively by introducing multiple factors. The core idea is to predict a continuous dependent variable (target variable) based on multiple independent variables (explanatory variables) by building a mathematical model. This model aims to quantify the linear relationship between independent variables and dependent variables in order to understand and predict the changing trend of the dependent variable. It is one of the widely used prediction tools in the field of data analysis, especially for processing data sets containing multiple explanatory variables.

[0067] The expression of the multiple linear regression model is: f(x) = k T x+b. Where x is the input vector, which contains multiple features (independent variables); f(x) is the output or response of the model (the predicted target variable); k T is the feature weight; b is the intercept or bias of the model. T and b make f(x) as close as possible to the true observed value y.

[0068] This example uses a multivariate linear regression model to perform linear regression on the generated new sample set to predict whether the patient has acute ischemic stroke. First, the sample set is randomly divided into a training set and a test set in a ratio of 8:2. The training set is used to build the model, while the test set is used to evaluate the performance of the model. In this scenario, we define the multivariate linear regression model as: f(x) = β0+β1x1+β2x2+...+β n x n . Where x=[x1,x2,...,x n ] represents the feature vector of each case data, f(x) represents the predicted probability of stroke, β0 is the intercept term of the model, β i (i=1,2,...,n) represents the feature weight. Next, we use the data in the training set to estimate the model parameters β0 and β i The goal is to make the predicted value f(x) as close as possible to the actual observed stroke prevalence. The least squares method is usually used to determine these parameters, that is, to minimize the sum of the squared differences between the predicted value and the actual value. Once the model parameters are determined, we can use the test set to evaluate the performance of the model. For each test sample, we input its feature vector x into the trained model to obtain the predicted value f(x). In order to evaluate the accuracy of the model, we can calculate the difference between the predicted result and the actual prevalence and use some common evaluation indicators, such as the mean square error (MSE) or the coefficient of determination R 2 To measure the quality of the model.

[0069] The unbalanced case data set processed by the algorithm of the present invention can significantly improve the prediction accuracy and overall balance of the model for minority class samples. Compared with the unprocessed data set, the new samples generated by SMOTE help reduce the bias of the model, enabling the model to better learn and reflect the characteristics of all samples, and fairly fit all categories of samples, especially pathological samples. Improving the robustness and generalization ability of the enhanced multivariate linear regression model, in practical applications, can more stably predict the disease progression of patients, identify risk factors related to the disease, reduce the risk of misdiagnosis and missed diagnosis, and provide more reliable support for clinical decision-making.

[0070] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all these improvements and changes should fall within the scope of protection of the appended claims of the present invention.

Claims

1. A method for processing imbalanced case data sets based on polar coordinates, characterized in that: The process comprises the following steps S110-S160: S110: Collect case data sets, where the case data includes healthy samples and pathological samples. Since the proportion of healthy sample data in the overall data set is relatively large, the healthy sample data are used as majority sample data, and the pathological sample data are used as minority sample data; the minority sample data and the majority sample data are respectively used to form a minority sample data set and a majority sample data set; S120: Perform PCA operation to generate a principal component analysis matrix; perform PCA operation without dimensionality reduction on the minority class sample data set to obtain a principal component analysis matrix; S130: Select neighboring samples and majority samples of the minority sample to generate polar coordinates; randomly select a sample from the minority class sample as a root sample, select its k neighboring samples and a majority sample, and calculate the polar coordinates of the main features; S140: Perform random sampling in the polar coordinate system to generate a new sample; randomly select a polar coordinate position in the defined annular range, and calculate and generate a feature vector of the new sample; S150: Convert back to the Cartesian coordinate system to generate a new sample matrix; convert the polar coordinate position into the Cartesian coordinate position, obtain the eigenvector of the new sample, and form a new principal component analysis matrix; S160: Add the generated new sample to the minority sample set and repeat the sampling; add the newly generated sample matrix to the minority sample set and repeat the above sampling steps until a predetermined number of sampling times is reached.

2. The method for processing imbalanced case data sets based on polar coordinates according to claim 1, characterized in that: The non-dimensionality reduction PCA operation in step S120 includes: S210 randomly selecting minority class samples as root samples; S220 performing non-dimensionality reduction PCA operation on matrix X to obtain a feature matrix P; S230 calculating a principal component analysis matrix and principal component eigenvectors.

3. The method for processing imbalanced case data sets based on polar coordinates according to claim 2, characterized in that: Let the minority sample set be D, the upsampling factor be n (repeated n times), and the number of sample neighbors be k; S210: Randomly select sample x from the minority class sample i As the root sample for generating new samples, i = 1, 2, ..., |D|; let the sample x including T features i The eigenvector of Then all samples are represented as matrices Where N is the number of all samples; S220: Perform a PCA operation on X without dimensionality reduction, that is, the length of the eigenvector is still T, and then obtain a eigenmatrix P; S230: Calculate the principal component analysis matrix X′ and the principal component eigenvector x i ′; let the principal component eigenvector in X′ be represented by x i ′; the formula is as follows: X′=XP.

4. The method for processing imbalanced case data sets based on polar coordinates according to claim 1, characterized in that: The process of generating polar coordinates in step S130 includes steps S310-S320; S310 selects neighboring samples, and S320 groups and calculates polar coordinates; S310: Select x in D i k nearest neighbor samples, let x ij For sample x i The jth nearest neighbor sample, j = 1, 2, ..., k; and select a nearest neighbor majority sample x o ,Right now S320: x' i The main features in are grouped in order, namely and The two-dimensional Cartesian space points that are the main features of the d-th group, where d = 1, 2, ..., T-1, and where and The two-dimensional Cartesian space point that is the main feature of the Tth group; similarly, x o The main eigenvector of is denoted as x o ', and x o ' are grouped in pairs in order; x o 'The two-dimensional Cartesian space points grouped in the polar coordinate system are used as the poles; a ray is drawn in the horizontal positive direction as the polar axis, and the positions of the two-dimensional Cartesian space points of all samples in the polar coordinate system are calculated; let the main features The polar coordinates are The polar coordinates are calculated as follows:

5. The method for processing imbalanced case data sets based on polar coordinates according to claim 1, characterized in that: The process of performing random sampling in the polar coordinate system to generate new samples in step S140 includes steps S410-S420; S410 defines a sample generation area, and S420 generates a new sample; S410 and S420: will be based on points and Point The minor arc with endpoints and Point The range enclosed by the minor arc of the endpoint is defined as the sampling range of the dth group of main features; the polar coordinates are randomly selected in this annular range The calculation expression is as follows: Where min(·,·) is the function for calculating the minimum value; λ1 and λ2 are random numbers between [0,1].

6. The method for processing imbalanced case data sets based on polar coordinates according to claim 1, characterized in that: Step S150 converts back to the Cartesian coordinate system and generates a new sample matrix in the following way: Let the main feature The polar coordinates are The Cartesian coordinates are then calculated as follows: Through calculation, the main features of the dth group in Cartesian coordinates are obtained as follows: and Repeat the above sampling steps d times to obtain the main feature vector x' after sampling new , forming the principal component analysis matrix X′ new ; Further, the sampling matrix X is obtained by inverse PCA calculation new , the calculation is shown as follows: X new =P -1 X′ new The sample matrix X new Add to the minority sample set D; repeat the above steps n times to obtain a new case sample data set D'.