A PSO Based Incremental Hyper-Sphere Partitioning Approach to Classification

Abstract

This paper proposes an incremental hyper-sphere partitioning approach for classification problems. Hyper-spheres that are close to the classification boundaries of a given problem are searched using an incremental approach based upon Particle Swarm Optimization (PSO). This new algorithm is proposed to tackle the difficulty of classification problems caused by the complex pattern relationship with a simplified expert rule structure. We solve classification problems through a combination of hyper-sphere partitioning and a Euclidean-distance based partitioning approach. Moreover, an incremental approach combined with output partitioning and pattern reduction is applied to cope with the curse of dimensionality. The algorithm is tested with seven datasets. The experimental results show that this proposed algorithm outperforms ILEGA (our former research work) and normal GA significantly in the final classification accuracy. In terms of the time complexity, it also gains significant improvement in comparison with ILEGA.