Multi-target particle swarm feature selection method and system based on chaotic population

Abstract

The invention discloses a multi-target particle swarm feature selection method and system based on a chaos population, and belongs to the technical field of data processing. The method comprises the following steps: acquiring a data set, and preprocessing the data set by using a chaos strategy; performing classification extraction on the preprocessed data by using a multi-target particle swarm algorithm; the feature correlation in the extracted data subset is analyzed, and the system comprises an initialization module, a feature extraction module and an updating module. According to the method and the system, the multi-target particle swarm optimization algorithm is optimized through the chaos strategy, the vitality of the population is improved, the classification error rate is effectively reduced, meanwhile, the number of features is remarkably reduced, a practical and effective solution is provided for solving the problem of high-dimensional data feature selection, and the method and the system have good application prospects. And challenges faced by high-dimensional data analysis are overcome successfully.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a multi-objective particle swarm feature selection method and system based on chaotic population. Background Art

[0002] Feature selection plays a vital role in machine learning and data mining. It is a method for directly selecting a subset of relevant features. By designing efficient and effective search strategies, feature selection methods can significantly reduce data dimensions and retain critical information. By eliminating these redundant and irrelevant features, the performance of subsequent data mining tasks can be significantly improved, including computational efficiency and learning effects. Reducing data dimensions can help improve the prediction accuracy of subsequent models, speed up training, improve model interpretability, and reduce the risk of overfitting, thus playing an important role in practical applications.

[0003] Chaotic systems are phenomena with random and irregular motion that appear in deterministic systems. Although their behavior seems disordered, they are actually driven by specific rules and show sensitivity to initial values. The inherent determinism of chaotic systems enables them to show a certain order or pattern, that is, the phenomenon of self-organization. By utilizing the randomness and ergodicity of chaotic mapping, introducing chaotic systems into optimization algorithms can help algorithms escape from local optimal solutions and search for global optimal solutions more effectively. By applying chaotic perturbations to individuals in a population, they can explore in a wider search space, increase population diversity, avoid falling into local optimality, and thus improve the overall performance of the algorithm. The application of chaotic systems in optimization algorithms provides a powerful tool for solving complex problems.

[0004] The multi-objective particle swarm optimization algorithm (MOPSO) introduces the Pareto dominance relationship into the PSO framework, stores non-dominated solutions through external archives, and updates and maintains them to achieve multi-objective optimization. MOPSO uses fitness value comparison to determine the position and velocity update of particles, and uses non-dominated sorting and crowding distance calculation to select particles. The algorithm optimizes the particle swarm through multiple iterations to generate multiple potential optimal solutions, helping decision makers to make trade-offs among multiple objectives. MOPSO effectively handles multi-objective problems, searches for and retains multi-directional excellent solutions, provides diversified solutions for complex problems, and helps understand and weigh the relationships between different objectives.

[0005] In recent years, more and more studies have applied chaotic systems and multi-objective particle swarm optimization algorithms to the field of feature selection. Multi-objective particle swarm optimization algorithms have been successful in improving classification performance, contributing from feature sorting to feature selection. Considering the importance of particle swarms in optimizing model performance, reducing redundant information, guiding feature selection, and model interpretability. However, in high-dimensional data environments, only considering local optimality is usually not enough. Therefore, a comprehensive framework is proposed to optimize the multi-objective particle swarm optimization algorithm. Summary of the invention

[0006] The purpose of the present invention is to provide a multi-objective particle swarm feature selection method and system based on chaotic population, which uses chaotic strategy to optimize the multi-objective particle swarm optimization algorithm, reduce the classification error rate and the number of feature subsets, and enhance data processing capabilities in the challenge of high-dimensional data dimensionality reduction, so as to achieve effective extraction of key information in high-dimensional data.

[0007] To achieve the above object, the present invention provides a multi-objective particle swarm feature selection method based on chaotic population, the steps comprising:

[0008] S1, obtain the data set and pre-process the data set using the chaos strategy;

[0009] S2, using multi-objective particle swarm algorithm to classify and extract the pre-processed data;

[0010] S3. Analyze the feature correlation in the extracted data subset.

[0011] Preferably, the step S1 includes:

[0012] S11, initialize the population and use TENT mapping to generate a series of chaotic variables;

[0013] S12, introduce chaos operator to generate new particles;

[0014] S13, using tent chaos mapping to perform chaos seeding, replacing one third of normal particles with the chaotic particles generated in step S12.

[0015] Preferably, the step S2 comprises:

[0016] S21, evaluating the fitness value and dominance relationship of each particle and establishing an archive set;

[0017] S22, update individual optimum and global optimum;

[0018] S23. Update the archive set and determine the global optimal solution.

[0019] Preferably, the step S22 includes:

[0020] Select the update formula according to the value of the environmental variable. If the environmental variable is 0, select the update formula based on differential evolution; if the environmental variable is 1, select the update formula based on genetic strategy;

[0021] Update the particle's position and velocity according to the selected formula;

[0022] Re-evaluate the fitness value and dominance relationship of each particle, and select the best individual and the best feature set in the current population based on the fitness value and dominance relationship.

[0023] Preferably, the step S23 includes:

[0024] The particles not selected in step S22 are stored in an archive set, and the particles in the archive set are compared. If the number of particles in the archive set exceeds a specified size, the redundant particles are deleted according to the crowding distance;

[0025] Judge the update of the global optimal solution. If it is not updated after one round of iteration, switch to another update formula.

[0026] The termination condition of the multi-objective particle swarm algorithm is judged. If it is satisfied, the optimal solution is output and the algorithm ends; otherwise, it continues to iterate until the global optimal solution is found.

[0027] Preferably, step S3 comprises:

[0028] S31: Calculate the correlation between features and classes;

[0029] S32: Calculate the correlation between features and classes;

[0030] S33: Update the global optimal solution according to the correlation degree and the correlation.

[0031] Preferably, the step S31 includes: calculating the mutual information between the feature and the class, and the magnitude of the mutual information value reflects the degree of correlation between the feature and the class label.

[0032] Preferably, the step S32 includes: using symmetric uncertainty to measure the correlation between features, normalizing the correlation, and obtaining the correlation between the features and the classes according to the correlation results.

[0033] Preferably, the step S33 includes:

[0034] Evaluate the correlation between the features in the current optimal subset and the class based on mutual information, and store the features with lower correlation in the archive set;

[0035] The symmetric uncertainty is used to measure the correlation between the features in the current global optimal feature subset and the features in the archive set, evaluate their mutual relationship, check the changes in the global optimal solution, and obtain the optimal data subset.

[0036] A multi-objective particle swarm feature selection system based on chaotic population includes an initialization module, a feature extraction module and an update module.

[0037] The initialization module is used to pre-process the high-dimensional data and generate a series of chaotic variables using TENT mapping; introduce chaos operators to generate new particles; use tent chaos mapping for chaos seeding and replace one-third of normal particles with newly generated chaotic particles;

[0038] The feature extraction module is used to extract the useful features in the high-dimensional data set and store the corresponding features and data; firstly, the fitness value and dominance relationship of each particle are evaluated to establish an archive set; then, the individual optimum and the global optimum are updated; finally, the archive set is updated to determine the global optimal solution;

[0039] The update module is used to analyze and update the stored data to obtain the optimal data subset. First, the correlation between the features and the classes is calculated; then the correlation between the features and the classes is calculated; finally, the global optimal solution is updated based on the correlation and correlation.

[0040] Therefore, the present invention adopts the above-mentioned multi-objective particle swarm feature selection method and system based on chaotic population, which has the following beneficial effects:

[0041] (1) Using Tent mapping to initialize the population enhances the diversity of the population;

[0042] (2) Introducing an environment recognition mechanism to dynamically select an appropriate population update strategy;

[0043] (3) The introduction of chaotic particles improves the vitality of the population, effectively reduces the classification error rate, and significantly reduces the number of features, providing a practical and effective solution to the problem of feature selection for high-dimensional data and successfully overcoming the challenges faced by high-dimensional data analysis.

[0044] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a flow chart of an embodiment of a multi-objective particle swarm feature selection method based on chaotic population of the present invention;

[0046] Figure 2 A schematic diagram of an embodiment of a multi-objective particle swarm feature selection system based on chaotic populations according to the present invention;

[0047] Figure 3 This is a comparison result diagram of an embodiment of a multi-objective particle swarm feature selection method based on chaotic population in the present invention. DETAILED DESCRIPTION

[0048] Example

[0049] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0050] Reference Figure 1 The present invention provides a multi-objective particle swarm feature selection method based on chaotic population, the steps comprising:

[0051] S1. Obtain a high-dimensional data set and use a chaotic strategy to preprocess the data set.

[0052] The preprocessing process includes:

[0053] S11. Use TENT mapping to generate a series of chaotic variables to initialize the population and obtain a set of variable sequences with chaotic characteristics. By using the chaotic variables generated by Tent mapping, the diversity and global search capabilities of the algorithm can be increased, which helps the optimization algorithm to better explore the search space and improve the efficiency and accuracy of problem solving.

[0054] S12. Introduce chaos operators to generate new particles, enhance the algorithm's search capability and avoid local optimal solutions. Chaotic operators have high randomness and variability, rarely generate similar particles, and have the opportunity to change in each dimension.

[0055] S13. Tent chaos mapping is used for chaos seeding. One third of the normal particles are replaced by the chaotic particles generated in step S12, which brings more diversity to the population initialization, helps the optimization algorithm to jump out of the local optimal solution, better explore the global search space, and accelerate the convergence speed.

[0056] S2. Use the multi-objective particle swarm algorithm to classify and extract the pre-processed data. The steps include:

[0057] S21. Evaluate the fitness value and dominance relationship of each particle and establish an archive set. According to the setting of the objective function, evaluate the performance of each particle in the solution space, determine the performance of each particle in the population by comparing the fitness value, and establish a dominance relationship. The archive set is used to store solutions with potential advantages, that is, particles with higher fitness values ​​or excellent performance on multiple objective functions.

[0058] S22, update individual optimal and global optimal. Select the update formula according to the value of the environmental variable. If the environmental variable is 0, select the update formula based on differential evolution; if the environmental variable is 1, select the update formula based on genetic strategy. According to the selected formula, update the position and speed of the particle; re-evaluate the fitness value and dominance relationship of each particle, and select the optimal individual and the best feature set in the current population based on the fitness value and dominance relationship. This step combines different update formulas and fitness evaluations to effectively guide the algorithm to find better solutions in the solution space, improving the performance and search efficiency of the algorithm.

[0059] S23, update the archive set and determine the global optimal solution. Store the particles that were not selected in the previous step in the archive set to retain potential high-quality solutions. Compare the particles in the archive set. If the number of particles in the archive set exceeds the specified size, delete the excess particles according to the crowding distance to maintain diversity. Judge the update of the global optimal solution. If it is not updated after one round of iteration, switch to another update formula, that is, if the previous environmental variable is 0, it becomes 1; vice versa. Judge the termination condition of the multi-objective particle swarm algorithm. If it is met, output the optimal solution and the algorithm ends; otherwise, continue to iterate until the global optimal solution is found. This step helps to find the best solution in the search space by maintaining the update of the archive set and the global optimal solution, and switching the update formula according to the specific situation, and ensures that the algorithm can output the best solution under the termination condition.

[0060] S3. Analyze the feature correlation in the extracted data subset, the steps include:

[0061] S31: Calculate the correlation between the feature and the class. The correlation between the feature and the class label is expressed by calculating the mutual information between the feature and the class. The value of the mutual information reflects the degree of association between the feature and the class label. The larger the value, the stronger the correlation between the two.

[0062] S32: Calculate the correlation between features and classes. Symmetric uncertainty is used to measure the correlation between features. Symmetric uncertainty corrects the bias when using mutual information for feature selection, and normalizes the correlation. The correlation between features and classes is obtained based on the correlation results. Through symmetric uncertainty, the correlation between features is standardized to the range of [0,1]. When the value is 0, it means that the two features are completely independent, that is, there is no predictive ability for the value of one feature to the value of another feature. When the value is 1, it means that there is a complete correlation between the two features.

[0063] S33: Update the global optimal solution based on the degree of correlation and relevance. Evaluate the degree of correlation between the features in the current optimal subset and the class based on the mutual information, and store the features with lower correlation in the archive set. Use the symmetric uncertainty to measure the correlation between the features in the current global optimal feature subset and the features in the archive set, evaluate their mutual relationship, and check the changes in the global optimal solution to obtain the optimal data subset. This step ensures the relevance and quality of the feature subset, avoids falling into the local optimal solution, and guides the update process of the feature subset to obtain higher quality feature selection results.

[0064] Reference Figure 2 , a multi-objective particle swarm feature selection system based on chaotic population, including initialization module, feature extraction module and update module,

[0065] The initialization module is used to preprocess the high-dimensional data and transfer the preprocessed data to the feature extraction module. The implementation process includes:

[0066] TENT mapping is used to generate a series of chaotic variables; chaos operators are introduced to generate new particles; tent chaos mapping is used for chaos seeding to replace one third of normal particles with newly generated chaotic particles.

[0067] The feature extraction module is used to extract useful features from high-dimensional data sets and store the corresponding features and data. The implementation process includes:

[0068] First, the fitness value and dominance relationship of each particle are evaluated to establish an archive set; then the individual optimum and global optimum are updated; finally, the archive set is updated to determine the global optimal solution;

[0069] The update module is used to analyze and update the stored data to obtain the optimal data subset. The implementation steps include:

[0070] First, the correlation between features and classes is calculated; then the correlation between features and classes is calculated; finally, the global optimal solution is updated based on the correlation and correlation.

[0071] Since the embodiments of the system part correspond to the embodiments of the method part, please refer to the description of the embodiments of the method part for the embodiments of the system part, which will not be repeated here.

[0072] In order to verify the effectiveness of the method of the present invention, a Vehicle data set was obtained, and the method CUMUPSO of the present invention was compared with five methods including NSGA-II, MOPSO, SPEA-II, ENORA and MOABC based on error rate and size of feature subsets. The comparison results are shown in Figure 2. Figure 3As shown in the figure, it can be seen that the method of the present invention can select fewer feature subsets to represent the entire data set with higher accuracy.

[0073] The present invention also provides a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the multi-objective particle swarm feature selection method based on chaotic population and is implemented as described above.

[0074] The present invention also provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, the multi-objective particle swarm feature selection method based on chaotic population and is implemented as described above.

[0075] Therefore, the present invention adopts the above-mentioned multi-objective particle swarm feature selection method and system based on chaotic population, optimizes the multi-objective particle swarm optimization algorithm through chaotic strategy, improves the vitality of the population, effectively reduces the classification error rate, and significantly reduces the number of features, providing a practical and effective solution to the problem of high-dimensional data feature selection, and successfully overcomes the challenges faced by high-dimensional data analysis.

[0076] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A multi-objective particle swarm feature selection method based on chaotic population, characterized in that the steps include: S1, obtain the data set and pre-process the data set using the chaos strategy; S2, using multi-objective particle swarm algorithm to classify and extract the pre-processed data; S3. Analyze the feature correlation in the extracted data subset.

2. The method for selecting multi-objective particle swarm features based on chaotic population according to claim 1, characterized in that: The step S1 includes: S11, initialize the population and use TENT mapping to generate a series of chaotic variables; S12, introduce chaos operator to generate new particles; S13, using tent chaos mapping to perform chaos seeding, replacing one third of normal particles with the chaotic particles generated in step S12.

3. The method for selecting multi-objective particle swarm features based on chaotic population according to claim 1, characterized in that: The step S2 comprises: S21, evaluating the fitness value and dominance relationship of each particle and establishing an archive set; S22, update individual optimum and global optimum; S23. Update the archive set and determine the global optimal solution.

4. The method for selecting multi-objective particle swarm features based on chaotic population according to claim 3, characterized in that: The step S22 comprises: Select the update formula according to the value of the environmental variable. If the environmental variable is 0, select the update formula based on differential evolution; if the environmental variable is 1, select the update formula based on genetic strategy; Update the particle's position and velocity according to the selected formula; Re-evaluate the fitness value and dominance relationship of each particle, and select the best individual and the best feature set in the current population based on the fitness value and dominance relationship.

5. The method for selecting multi-objective particle swarm features based on chaotic population according to claim 4, characterized in that: The step S23 comprises: The particles not selected in step S22 are stored in an archive set, and the particles in the archive set are compared. If the number of particles in the archive set exceeds a specified size, the redundant particles are deleted according to the crowding distance; Judge the update of the global optimal solution. If it is not updated after one round of iteration, switch to another update formula. The termination condition of the multi-objective particle swarm algorithm is judged. If it is satisfied, the optimal solution is output and the algorithm ends; otherwise, it continues to iterate until the global optimal solution is found.

6. The method for selecting multi-objective particle swarm features based on chaotic population according to claim 1, characterized in that: The step S3 comprises: S31: Calculate the correlation between features and classes; S32: Calculate the correlation between features and classes; S33: Update the global optimal solution according to the correlation degree and the correlation.

7. The method for selecting multi-objective particle swarm features based on chaotic population according to claim 6, characterized in that: The step S31 includes: calculating the mutual information between the feature and the class, and the value of the mutual information reflects the degree of correlation between the feature and the class label.

8. The method for selecting multi-objective particle swarm features based on chaotic population according to claim 7, characterized in that: The step S32 includes: using symmetric uncertainty to measure the correlation between features, normalizing the correlation, and obtaining the correlation between the features and the classes according to the correlation results.

9. The method for selecting multi-objective particle swarm features based on chaotic population according to claim 8, characterized in that: The step S33 comprises: Evaluate the correlation between the features in the current optimal subset and the class based on mutual information, and store the features with lower correlation in the archive set; The symmetric uncertainty is used to measure the correlation between the features in the current global optimal feature subset and the features in the archive set, evaluate their mutual relationship, check the changes in the global optimal solution, and obtain the optimal data subset.

10. A multi-objective particle swarm feature selection system based on chaotic population, using a multi-objective particle swarm feature selection method based on chaotic population according to any one of claims 1 to 9, characterized in that: Including initialization module, feature extraction module and update module, The initialization module is used to pre-process the high-dimensional data and generate a series of chaotic variables using TENT mapping; introduce chaos operators to generate new particles; use tent chaos mapping for chaos seeding and replace one-third of normal particles with newly generated chaotic particles; The feature extraction module is used to extract the useful features in the high-dimensional data set and store the corresponding features and data; firstly, the fitness value and dominance relationship of each particle are evaluated to establish an archive set; then, the individual optimum and the global optimum are updated; finally, the archive set is updated to determine the global optimal solution; The update module is used to analyze and update the stored data to obtain the optimal data subset. First, the correlation between the features and the classes is calculated; then the correlation between the features and the classes is calculated; finally, the global optimal solution is updated based on the correlation and correlation.

Images