Manifold distance-based particle swarm optimization for classification
Ruochen Liu, Lixia Wang, Pei Wu, Xingtong He, Licheng Jiao · Transactions of the Institute of Measurement and Control · 2013
Classification is one of the most important research topics in data mining. In this paper, a manifold distance-based particle swarm optimization (MDPSO) for classification is proposed. Firstly, a new selection strategy is designed to maintain information of some good solutions as well as enhance the diversity of swarm. Furthermore, a manifold distance is utilized to measure the similarity of two samples, which measures the geodesic distance along the manifold and can better reflect the global consistency of the given problems. MDPSO is applied to solve 10 benchmark classification problems selected from UCI (University of California Irvine) datasets and 10 artificial data sets. The experimental results show that MDPSO outperforms the typical C4.5, nearest neighbour, and Michigan particle swarm optimization on most of the testing problems used in this paper. In addition, MDPSO is applied to a real-world application problem, namely handwritten digit recognition, with a good performance obtained.