基于改进飞蛾火焰算法同步优化SVM参数和特征选择

Simultaneous Optimization of SVM Parameters and Feature Selection Based on Improved Moth-Flame Algorithm

摘要: 特征选择和参数优化都是提高机器学习分类正确率和效率的重要方法，本文提出了一种基于改进的飞蛾火焰算法(IMFO)同步优化支持向量机(SVM)参数和特征选择的方法。针对飞蛾火焰算法(MFO)寻优精度较低和容易陷入局部最优的问题，首先，利用反馈共享机制增强飞蛾之间的信息交流，使个体容易脱离局部最优。其次，引入惯性权重因子改进飞蛾位置更新公式，增强算法的勘探能力。最后，将IMFO用于同步优化SVM的参数和特征选择中，并在12个UCI数据集上进行了特征选择实验，实验结果表明，该方法能有效地优化SVM参数和特征子集，在提高分类准确率的同时，减少特征数量。

Abstract: Both feature selection and parameter optimization are important methods to improve the accuracy and efficiency of machine learning classification. In this paper, a method based on Improved Moth-Flame Algorithm (IMFO) is proposed to synchronously optimize support vector machine (SVM) parameters and feature selection. In order to solve the problem of low optimization accuracy and easily fall into local optimization of the Moth-Flame Algorithm (MFO), firstly, a feedback-sharing mechanism is used to enhance the information exchange between moths, so that individuals can easily escape from local optimization. Secondly, the inertia weight factor is introduced to improve the moths’ position updating formula to enhance the exploration ability of the algorithm. Finally, IMFO is used for synchronous SVM parameters and feature selection, and feature selection experi-ments are carried out on 12 UCI datasets. The experimental results show that the proposed method can effectively optimize SVM parameters and feature subsets, improve classification accuracy and reduce the number of features.