An efficient algorithm for initializing centroids in K-means clustering

Clustering represents one of the most popular knowledge extraction algorithms in data mining techniques. Hierarchical and partitioning approaches are widely used in this field. Each has its own advantages, drawbacks and goals. K-means represents the most popular partitioning clustering technique, however it suffers from two major drawbacks; time complexity and its sensitivity to the initial centroid values. The work in this paper presents an approach for estimating the starting initial centroids throughout three process including density based, normalization and smoothing ideas. The proposed algorithm has a strong mathematical foundation. The proposed approach was tested using a free standard data (20000 records). The results showed that the approach has better complexity and ensures the clustering convergence.