Comprehensible Predictive System Model using Parameter less Fast K-Means (EPFK-Means) for web usage data

Predictive modeling approaches provide a way to streamline operational business processes. The model proposed here consists of two major steps, namely, preprocessing and prediction. We suggest a clustering method to cluster pages in a session with respect to the number of visits. Here, the conventional K-Means algorithm is modified in three manners. The first is to automatically find K (number of clusters) value using ensemble method. The second is to automatically identify initial centroids. The third is to propose method that can reduce the number of distance calculations performed in order to reduce its complexity. This clustering algorithm is combined with longest common sequence algorithm to identify the future needs of the user. Experimental results prove that the inclusion of clustering algorithm improves the prediction performance and the prediction process as against longest common sequence algorithm and conventional K-Means algorithm.