K-means clustering versus validation measures: a data distribution perspective