作者: Hoai Bach Nguyen , Bing Xue , Peter Andreae
DOI: 10.1007/978-3-319-31204-0_46
关键词: Multi-swarm optimization 、 Computer science 、 Overfitting 、 Feature selection 、 k-nearest neighbors algorithm 、 Particle swarm optimization 、 Artificial intelligence 、 Mutual information 、 Pattern recognition 、 Metaheuristic 、 Fitness function
摘要: Feature selection is a pre-processing step in classification, which selects small set of important features to improve the classification performance and efficiency. Mutual information very popular feature because it able detect non-linear relationship between features. However existing mutual approaches only consider two-way interaction In addition, most methods, calculated by counting approach, may lead an inaccurate results. This paper proposes filter algorithm based on particle swarm optimization (PSO) named PSOMIE, employs novel fitness function using nearest neighbor estimation (NNE) measure quality set. PSOMIE compared with all two traditional approaches. The experiment results show that successfully guides PSO search for number while maintaining or improving over methods. provides strong consistency training test results, be used avoid overfitting problem.