Efficient Fuzzy C-means Based Reduced Feature Set Association Rule Mining Approach for Predicting the User Behavioral Pattern in Web Usage Mining

Abstract

Online business and marketing are becoming popular now a day due to the wide variety of products available from multiple vendors in online. One of the major challenges of e-business merchants is predicting the buying and selling patterns of online customers. Global level competition is another challenge faced by online merchants due to the lowest prices and offers provided by multiple sellers for the same or similar product. Hence, the development of an efficient web mining framework to analyze and predict buyer’s interest based on the browsing history will be a great support to the online sellers by providing exact or relevant product details to the buyers in online. Association rule mining plays an essential role in Web Mining for finding the most frequent and predictive patterns of the user. The major challenge in this approach is the generation of many rules for a huge volume of datasets. Decision making based on association rule mining is critical because knowledge is not directly present in frequent patterns. This research work focuses on the analysis of standard web mining approaches such as k-means clustering, fuzzy c-means clustering, fuzzy k-medoids clustering and fuzzy clustering with weighted session page matrix approach. In this work, MSNBC dataset from UCI Machine Learning Repository has been taken for analysis. Dimensionality reduction plays an important role in the accurate classification of users with respect to their interests. This research work proposed fuzzy C-Means using Kernel Principal Component Analysis (k-PCA) as a dimensionality reduction method based association rule mining classification, grouping and pattern prediction with 100% “confidence” along with a “lift” value greater than 1. The “support” value also shows higher compare with other existing methods and features are effectively reduced in the proposed architecture.