Subset K-Means Approach for Handling Imbalanced-Distributed Data

Abstract

The effectiveness of clustering analysis relies not only on the assumption of cluster number but also on the class distribution of the data employed. This paper represents another step in overcoming a drawback of K-means, its lack of defense against imbalance data distribution. K-means is a partitional clustering technique that is well-known and widely used for its low computational cost. However, the performance of k-means algorithm tends to be affected by skewed data distributions, i.e., imbalanced data. They often produce clusters of relatively uniform sizes, even if input data have varied cluster size, which is called the “uniform effect.” In this paper, we analyze the causes of this effect and illustrate that it probably occurs more in the k-means clustering process. As the minority class decreases in size, the “uniform effect” becomes evident. To prevent the effect of the “uniform effect”, we revisit the well-known K-means algorithm and provide a general method to properly cluster imbalance distributed data.

The proposed algorithm consists of a novel under random subset generation technique implemented by defining number of subsets depending upon the unique properties of the dataset. We conduct experiments using ten UCI datasets from various application domains using five algorithms for comparison on eight evaluation metrics. Experiment results show that our proposed approach has several distinctive advantages over the original k-means and other clustering methods.