Feature selection screening method based on enhanced binary bat algorithm

Through the feature selection method based on the enhanced binary bat algorithm, global search and diversity enhancement technology are used to solve the problem of useless features in machine learning algorithm training, improve the efficiency and accuracy of feature selection, and achieve the performance improvement of machine learning algorithms .

CN119046655APending Publication Date: 2024-11-29CHANGCHUN VOCATIONAL INST OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410726133.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-06-06
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

The training process of machine learning algorithms is time-consuming and computationally expensive, and there are a large number of useless and redundant features in the training dataset, which affects the effectiveness of their classification, clustering and regression. Effective feature selection methods are needed to improve performance and reduce training. Consumption.

Method used

The feature selection screening method based on the enhanced binary bat algorithm is adopted. By introducing the global search of Levi flight, population diversity enhancement, chaos-based loudness method and V-shaped binary operator, the feature selection problem is transformed into a fitness function, and the Feature selection filter. Specific steps include data set construction, random generation of populations, updating bat location and velocity, generating new solutions, updating pulse volume and pulse rate, and sorting fitness values ​​to find the optimal solution.

Benefits of technology

Effectively eliminate useless and redundant features in the dataset, reduce the number of features, improve the classification accuracy of machine learning algorithms, and reduce training consumption. It is suitable for feature selection for NP difficult problems.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119046655A_ABST
    Figure CN119046655A_ABST
Patent Text Reader

Abstract

The invention discloses a feature selection screening method based on an enhanced binary bat algorithm, and the method comprises the steps: carrying out the modeling of a feature selection problem, and converting the feature selection problem into a fitness function; a global search method of Levy flight, a population diversity enhancement method, a loudness method based on chaos and a V-shaped binary operator are introduced, so that the method is more suitable for a feature selection problem, and a feature selection screening method is realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of computer data analysis. Background Art

[0002] With the development of communication networks, computers, and artificial intelligence technology, a large amount of data is generated, stored, and utilized, and the amount of data is growing exponentially. Machine learning can effectively utilize this data and learn rules and patterns from it to help people make predictions and decisions. Machine learning algorithms have been successfully applied to various fields of life, such as medicine, materials science, and physics. Specifically, machine learning algorithms can extract features from various types of data and use these features to train models for classification, regression, and clustering operations.

[0003] Despite the advantages of the above machine learning algorithms in terms of effectiveness, applicability, and scalability, there are still some challenges and urgent problems that need to be solved in machine learning algorithms. First, the training of machine learning algorithms is a time-consuming and computationally intensive process, which may lead to limited application of machine learning algorithms; second, the training data sets of machine learning algorithms are derived from the features they extract, which are mostly extracted using automated tools and human experience, and have many repeated, meaningless, and misleading features, which further slow down the training process of machine learning algorithms and reduce the effectiveness of machine learning algorithms in classification, clustering, and regression. Therefore, eliminating these useless and redundant features is very important to improve the performance of machine learning algorithms and reduce their training consumption.

[0004] Feature selection is an effective way to solve the above problems. Feature selection can eliminate useless and redundant features in the data set, thereby reducing the number of features and improving the classification accuracy of the machine learning algorithm. For a data set with N features, there are 2 N feature selection schemes are available, resulting in a combinatorial explosion. Therefore, selecting a feature subset with high classification accuracy and a small number of features can be viewed as an optimization problem. On the other hand, the feature selection problem has also been proven to be an NP-hard problem. It is of great significance to select and propose a suitable algorithm to solve the feature selection problem. Summary of the invention

[0005] The present invention intends to disclose a feature selection screening method based on an enhanced binary bat algorithm, modeling the feature selection problem and then converting it into a fitness function. By introducing the global search method of Levy flight, the population diversity enhancement method, the chaos-based loudness method and the V-shaped binary operator to make it more suitable for the feature selection problem, the feature selection screening method is realized. The steps of the present invention are:

[0006] (1) Dataset construction: collect ten machine learning datasets;

[0007] (2) Randomly generate populations;

[0008] (3) Update the bat’s position and velocity;

[0009] (4) Generate a new solution: Generate a uniformly distributed random number N between 0 and 1 rand , if N rand <r i , use the suboptimal individuals to guide the population, otherwise select the optimal individuals to guide the population. Finally, generate new individuals around the selected individuals;

[0010] (5) accept new individuals and update the pulse volume and pulse rate;

[0011] (6) Sort the fitness values ​​of all bats and find the current optimal solution and optimal value.

[0012] Description of the accompanying tables and graphs

[0013] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0014] Figure 1 It is a schematic diagram of the feature screening process of the present invention;

[0015] Figure 2 is the convergence speed obtained by different benchmark algorithms in the embodiments of the present invention;

[0016] Figure 3 is the convergence speed obtained by different bat algorithms in the embodiments of the present invention; DETAILED DESCRIPTION

[0017] The specific implementation steps of the present invention are:

[0018] (1) Dataset Construction

[0019] (2) Randomly generate population

[0020] a) Define relevant parameters: Randomly generate a population P = [X1, X2, ..., X Npop ], where N pop is the population size. At the same time, define the bat dimension N dim , maximum iteration t max , fitness function, etc.

[0021] b) Initialization: Initialize bat population, pulse frequency, pulse rate and loudness;

[0022] c) Calculate the fitness values ​​of all bats.

[0023] (3) Update the bat's position and speed

[0024] By adjusting the frequency to generate new bats, the process is as follows:

[0025] a) Using the formula f i =f min +(f max -f min )×β update frequency;

[0026] b) Using the formula Model the speed of the i-th bat;

[0027] c) Using the formula Update the jth dimension of the position of the i-th bat. Introduce the binary operator: Among them, v i,j is the j-th dimension of the velocity of the i-th bat.

[0028] (4) Generate new solutions

[0029] The best bat in the group is searched locally with a certain probability, new solutions are generated through random perturbations, and out-of-bounds processing is performed on the new solutions to enhance the search capability.

[0030] a) Using formula X N =X * +ε×A t Perform local search. Among them, X N is a newly generated bat after random walk, A t is the loudness of all bats in the t iteration, ε is a random variable with a range of [-1, 1];

[0031] b) Introduce the Levy flight global search method to improve the algorithm's search ability. Since Levy flight is a random walk and a heavy-tailed probability distribution, the search range can be expanded by alternating between short-distance and long-distance searches. Therefore, in each iteration, a new individual is generated based on the best individual and Levy flight. It can be expressed as: Where α is a parameter, and the levy distribution is expressed as: levy(λ)~u=t -λ (1<λ<3);

[0032] c) Introducing a population diversity enhancement method, using suboptimal bat X ** Spawn new bat: X N2 =X ** +εA t ;

[0033] d) A chaos-based loudness method is introduced, using a new one-dimensional fractional chaotic map to update the loudness: In the formula, C t is the tth dimension of the one-dimensional fractional chaotic map, which can be expressed as:

[0034] e) Introduce binary operators and use V-shaped functions:

[0035] (5) Update pulse volume and pulse rate

[0036] a) Using the formula Update pulse volume and pulse rate;

[0037] b) Using the formula The optimal solution and suboptimal solution are used to guide the population at the same time to enhance the population diversity of the algorithm.

[0038] (6) Sort the fitness values ​​of all bats and find the current optimal solution and optimal value.

[0039] Example

[0040] This paper takes 10 typical UCI datasets as examples, and the main information is shown in Table 1:

[0041] Table 1 Benchmark datasets

[0042]

[0043] Table 2 Key parameters of different algorithms

[0044]

[0045] Table 3 Optimization results obtained by different algorithms (the best value is highlighted)

[0046]

[0047] Table 4 Optimization results of different bat algorithms (the best value is highlighted)

[0048]

[0049] Table 5 Optimization results of different algorithms under the decision tree (the best value is highlighted)

[0050]

[0051] Table 6 Optimization results of different algorithms under random forest (the best value is highlighted)

[0052]

[0053] In addition, the CPU used is the 11th generation Intel(R) Core(TM) i7-11700@2.50GHz, and the RAM is 16GB. Python is used to implement the simulation code, and KNN (k=5) is adopted. In addition, a and b in the fitness function are set to 0.99 and 0.01, respectively. This paper also introduces binary PSO (BPSO), BGWO, binary DA (BDA), and binary BA (BBA) as benchmarks, and the key parameters of these algorithms are shown in Table 2. Among them, the overall scale and number of iterations of EBBA and other benchmarks are set to 24 and 100, respectively. In addition, in order to avoid random deviations in the experiment, each algorithm is independently executed 30 times in these selected datasets according to the provisions of the central limit theorem. In addition, 80% of the instances are used for teaching, and the remaining 20% ​​are used for testing.

Claims

1. A feature selection method based on Levy flight enhanced binary bat algorithm, comprising the following steps: Construction of dataset; Randomly generate a population P = [X1, X2, ..., X Npop ], where N pop is the population size. At the same time, define the bat dimension N dim , maximum iteration t max , fitness function, etc. Initialize the bat population, pulse frequency, pulse rate and loudness. Calculate the fitness value of all bats; Generate new bats by adjusting the frequency: f i =f min +(f max -f min )×βFormula 1; In formula 1, f max and f min are the upper and lower limits of all bat frequencies, respectively, and β is a random number between [0,1]; In formula 2, X * is the bat with the highest fitness function value in the group, and the speed of the i-th bat is modeled; In formula 3, the jth dimension of the position of the i-th bat is updated. Introducing the binary operator: in, v i,j is the jth dimension of the velocity of the i-th bat; The best bat in the group is searched locally with a certain probability, new solutions are generated through random perturbations, and the new solutions are processed out of bounds to enhance the search capability. X N =X * +ε×A t Formula 4; In formula 4, X N is a newly generated bat after random walk, A t is the loudness of all bats in the t iteration, ε is a random variable with a range of [-1, 1]; The Levy flight global search method is introduced to improve the algorithm's search ability. Since the Levy flight is a random walk and a heavy-tailed probability distribution, the search range can be expanded by alternating between short-distance and long-distance searches. Therefore, in each iteration, a new individual is generated based on the best individual and the Levy flight; In formula 5, α is a parameter, and the levy distribution is expressed as: levy(λ)~u=t -λ (1<λ<3); Update the pulse volume and pulse rate according to Equation 6 and Equation 7: In formula 6, is the loudness of all bats in t iterations, α is a parameter ranging from 0 to 1, is the pulse rate updated as the iteration process proceeds, γ>0 is a parameter.

2. The feature selection method according to claim 1, characterized in that: The population size and number of iterations are set to 24 and 100 respectively.