Hybrid K-Means and Improved Self-Adaptive Particle Swarm Optimization for Data Clustering

Abstract

Data Clustering has become an important mechanism for data exploration and understanding. K-Means algorithm is currently one of the most popular clustering techniques, due to its simplicity and scalability. However, K-Means performance is highly influenced by the choice of initial cluster centers, which may lead to suboptimal solutions. In this paper, a novel hybrid partitional clustering algorithm is proposed, named IDKPSOC-k-means, based on an improved self-adaptive Particle Swarm Optimization (PSO) and K-Means, which uses a crossover operator to improve PSO capability to escape from local minima points from the problem space. To evaluate the performance of the proposed approach, experiments have been performed on sixteen benchmark data sets obtained from UCI Machine Learning Repository. The experimental evaluation, conducted by the use of Friedman hypothesis tests in relation to four clustering metrics, has shown the effectivity of the proposed model in relation to the comparison algorithms.