A new hybrid clustering approach based on heuristic Kalman algorithm

Abstract

Clustering is an important methodology for data mining and data analysis. K-Means is a simple and fast algorithm for clustering data. However the performance of K-means is highly sensitive on the initial seed of the algorithm. Heuristic Kalman Algorithm (HKA) is a population based stochastic optimization technique which is an effective method for searching a near-optimal solution of a function. Although HKA has good global search characteristics, it is shown that when directly applied on clustering it performs poorly. This paper proposes a new approach KHKA, which combines the benefits of the global nature of HKA and the fast convergence of K-means. KHKA was implemented and benchmarked on synthetic and real datasets from UCI Machine Learning Repository. The results were compared with other population based, stochastic algorithms. Results show that KHKA is a promising algorithm and was able to perform better than the compared algorithms with respect to the used datasets.