Exploring the Role of Multi-Agent Systems in Improving K-Means Clustering Method

Abstract

Clustering algorithms are attracting much application interest due to the significant growth in the rate of data generation. However, the high computational complexity of the existing clustering algorithms has rendered them ineffective in several ways. Therefore, this study proposes a novel parallel clustering algorithm that integrates Multi-Agent Systems (MAS) with the K-means algorithm; hence, the proposed novel algorithm is termed Multi-agent-K-means (MK-means). The MK-means is based on the concept of separating the activating agents to perform clustering while considering a different subset of features for each agent. This is aimed at the preservation of the dataset while improving the clustering accuracy. The cluster centers are calculated for each partition before being merged and clustered again. The statistical significance of the proposed approach is provided. The effectiveness of the proposed MK-means clustering algorithm was evaluated using Wine datasets, and the performance is compared with that of the traditional K-means algorithm. The results analysis proved the superiority of the proposed MK-mean clustering algorithm over the traditional version of the K-means algorithm. This suggests the suitability of the proposed algorithm as a clustering technique that can also be applied to other clustering algorithms that are based on the concept of initial cluster centroids selection.