The Feature Selection Based On Adaptive Particle Swarm Optimization

In the pattern classification,the data often have many irrelevant or redundant features,thus affecting the accuracy of classification.Feature selection,as an effective means to solve this problem,has always been a hotspot in machine learning.With the increase of data size,the original feature selection method has not met the requirements.Feature selection can be regarded as a process of dynamic optimization,and the particle swarm optimization algorithm is a popular algorithm of swarm intelligence algorithm,receiving wide attention with its simplicity,accessibility and high optimization efficiency.The combination of particle swarm optimization algorithm and feature selection method has also become a research hotspot.A large number of studies have shown that based on particle swarm optimization algorithm combined with feature selection is feasible,and has a good performance.In this paper,we focus on the improvement of particle swarm optimization and the combination of feature selection and particle swarm optimization.Firstly,the particle swarm optimization algorithm is improved.Ordinary particle swarm optimization is easy to fall into the local optimum because of its limitation.On the basis of bare bones particle swarm algorithm,an adaptive algorithm based on interference factor is proposed.In the initial process of the algorithm,the chaotic model is introduced to increase the diversity of the initial particles.An adaptive factor,meanwhile,is introduced in the updating mechanism to increase global searching ability and improve classification accuracy.Secondly,particle's local and global optimal iteration formula is improved.The number of features is introduced in the updating process,as well as the mutual information in the process of decoding to filter the features.The purpose of feature selection is to achieve the best optimization using least amount of features.In previous research,the number of features in the feature subset is neglected for better classification effect.Finally,a feature selection algorithm based on hybrid pattern evaluation is proposed.The feature selection process is divided into two phases,the first phase is filter model based on rough set,the second phase is wrapper model based KNN algorithm.In order to validate the proposed theory,we carried out classification experiments on different types of data sets and selecteddifferent classification models for validation.Experimental results verify the effectiveness and practicability of the algorithm.