Simultaneous SVM Parameters and Feature Selection Optimization Based on Improved Slime Mould Algorithm

Abstract

To address the problems of low classification accuracy, redundancy of feature subsets, and performance susceptibility to parameters in wrapper-based feature selection in traditional Support Vector Machine (SVM), an improved Slime Mould Algorithm (ISMA) was proposed for simultaneous optimization of SVM parameters and feature selection. Firstly, the golden section coefficient was introduced to improve the position update mechanism of slime mould individuals, so as to accelerate the convergence speed of SMA and improve the local development ability and convergence accuracy of SMA. Secondly, an adaptive lens-imaging learning strategy was proposed, which selects a solution with the largest difference from the optimal solution and the highest fitness value through the Fitness-Distance Balance method, and reverse learning was only performed on its specific dimension, in order to better balance the exploration and development capabilities of SMA. Finally, the vertical crossover was used to expand the search range, thereby reducing the probability of the algorithm falling into the local optimum. The experimental results of the performance test of the improved algorithm show that ISMA has higher solution accuracy, better stability and faster convergence speed, and has high performance in practical engineering optimization problems. Use ISMA to optimize SVM parameters and feature selection simultaneously. Simulation experiments of feature selection were carried out on 10 UCI datasets and the experimental results show that the simultaneous feature selection optimization method proposed in this paper can obtain higher classification accuracy under the condition of effectively reducing the feature dimension, and the classification accuracy on 7 datasets is as high as 90% above, which reached 100% on 2 datasets. In order to further prove the practicability of this method in practical problems, the simultaneous feature selection optimization method based on ISMA is applied to the microarray gene expression classification problem, and the experimental results on the two cancer datasets show that the proposed method obtains high classification accuracy when using a small number of predicted genes, and has good application value in cancer diagnosis and classification.

Topics & Concepts