A novel K-means and K-medoids algorithms for clustering non-spherical-shape clusters non-sensitive to outliers

Determination of the optimal number of clusters, the random selection of the initial centers, the non-detection of non-spherical clusters, and the negative impact of outliers are the main challenges of the K-means algorithm. In this paper, to tackle these issues three simple and intelligent algorithms are proposed by changing the structure of the K-means and K-medoids algorithms. The difference between these algorithms is in the selection of the initial centers and the stop condition. A method has been proposed to obtain the overlap space between the clusters. Using this method, a modified K-means algorithm is developed for the clustering of non-spherical data. These algorithms are designed in a way that they are not sensitive to outliers and can identify clusters having non-spherical shapes. The performance of the proposed methods is illustrated by applying the proposed algorithms to the different data sets and by comparing the results of the algorithms with other methods.