Genetic Algorithm-Based Optimization of Clustering Data Points by Propagating Probabilities

Abstract

Clustering is among the pivotal elementary operations in the field of data analysis. The efficiency of a clustering algorithm depends on a variety of factors like initialization of cluster centers, shape of clusters, density of the dataset, and complexity of the clustering mechanism. Previous work in clustering has managed to achieve great results but at the expense of a trial and error approach to achieve optimal values for user-defined parameters which have a huge bearing on the quality of the clusters formed. In this work, we propose a solution that optimizes the user-defined parameters for clustering algorithm called Probability Propagation (PP) by harnessing the capabilities of Genetic Algorithm (GA). In order to overcome this sensitivity in PP, a novel optimization technique is applied by obtaining the optimal values of $\delta$ and s using GA by maximizing inter-cluster spread and minimizing intra-cluster spread among the clusters being formed. The proposed method was found to retrieve top chromosomes (bandwidth and s) with a similar number of clusters, thus eliminating the sensitivity of user-defined parameters which is optimized by using GA.