Robust clustering in the presence of outliers using a Majorization-Minimization Kmeans Algorithm

The Kmeans clustering algorithm is highly regarded as one of the most common statistical techniques utilized in unsupervised analysis. However, its effectiveness can be compromised by the presence of outliers. In this work, we propose a robust clustering method in the presence of outliers. Our approach is a hybrid of the Kmeans algorithm and the majorization-minimization technique. Alongside investigating the theoretical foundations of our approach, we evaluate its efficacy using numerical experiments. Based on a simulation study, we observed that the proposed algorithm is robust against outliers. Also, applications of the algorithm on clustering of four real-world datasets, which contain outliers, are presented. Our experimental results demonstrate superior performance of the proposed algorithm, achieving significant improvements of criteria such as Purity, Normalized mutual information, Accuracy, Specificity and Sensitivity compared to existing clustering techniques.