Evaluation of Web Session Cluster Quality Based on Access-Time Dissimilarity and Evolutionary Algorithms

摘要

Web session cluster refinement is one of the major research issues for the improvement of cluster quality in recent days. The motive of refinement using Evolutionary Algorithms is quite obvious because in any clustering algorithm the obtained clusters shall have some data items that are inappropriately clustered, hence, never giving us well separated and cohesive clusters. Hence the quality of clusters is improved using refinement techniques. Initial clusters are formed using K-Means clustering algorithm which suffers from local minima problem. The refinement on clusters is performed on the basis of access and time features (Modified Knockout Refinement Algorithm) which is a distance based dissimilarity, Genetic Algorithm (GA), Particle Swarm Optimization (PSO) and a combination of MKRA with GA and MKRA with PSO. Results are evaluated on five synthetic datasets and three real datasets. Further, it is shown experimentally that effectiveness of combining MKRA with evolutionary techniques produces better quality clusters.