Hybrid K-Means and Improved Group Search optimization Methods for Data Clustering
Luciano D. S. Pacífico, Teresa B. Ludermir · 2018
Data clustering has become an important tool for exploratory data analysis and knowledge discovery in data sets, finding applications in many fields, such as engineering, medicine, data mining, social sciences, and so on. Since traditional clustering techniques such as K-Means are local optimizers, the quality of the final solutions achieved by such approaches may be very poor, once these methods are quite sensible to local minima points. In this work, we propose two new hybrid partitional evolutionary algorithms for data clustering which use Group Search optimization (GSO) as a global searcher to execute clustering, where K-Means is employed as a local searcher to complement the search performed by GSO. Also, a selection operator is employed to prune the worst individuals from GSO population. We compare the proposed approaches by employing some hybrid evolutionary and K-Means algorithms from data clustering literature. The experimental results show that the proposed hybrid methods are able to achieve better performances than the comparison approaches in an overall evaluation obtained through Friedman/Nemenyi statistical analysis in relation to fifteen selected real-world benchmark problems from UCI Machine Learning Repository.