Overlapping clustering with k-median extension algorithm: An effective approach for overlapping clustering

Abstract

Most natural world data involves overlapping communities where an object may belong to one or more clusters, referred to as overlapping clustering. However, it is worth mentioning that these algorithms have a significant drawback. Since some of the algorithm uses k-means, it also inherits the characteristics of being noise sensitive due to the arithmetic mean value which noisy data can considerably influence and affects the clustering algorithm by biasing the structure of obtained clusters. This paper proposed a new overlapping clustering algorithm named OCKMEx, which uses k-median to identify overlapping clusters in the presence of outliers. This new method aims to determine the insensitivity of the OCKMEx algorithm in locating data points that overlap even with outliers. An experimental evaluation of the algorithm was conducted wherein synthetic datasets served as a data source, and the F1 measure criterion was applied to assess the OCKMEx algorithm performance. Results indicate that the OCKMEx algorithm implementing the use of k-median performed a higher accuracy rate of 100% in identifying data points that overlap even with outliers compared to the existing k-means algorithm. The algorithm exhibited a promising performance in identifying overlapping clusters and was resistant to outliers.