Incremental k-Means Method

In the last few decades, k-means has evolved as one of the most prominent data analysis method used by the researchers. However, proper selection of k number of centroids is essential for acquiring a good quality of clusters which is difficult to ascertain when the value of k is high. To overcome the initialization problem of k-means method, we propose an incremental k-means clustering method that improves the quality of the clusters in terms of reducing the Sum of Squared Error (SSEtotal). Comprehensive experimentation in comparison to traditional k-means and its newer versions is performed to evaluate the performance of the proposed method on synthetically generated datasets and some real-world datasets. Our experiments shows that the proposed method gives a much better result when compared to its counterparts.