Optimal Association Rule Mining for Web Page Prediction using Hybrid Heuristic Trained Neural Network

Abstract

Today, data mining, which is a branch of web mining acts a fundamental role in diverse applications like health care data extraction, education system, search engines for evaluating their performance rank over other systems. Web Page prediction (WPP) is a classification issue in which the prediction of web pages is accomplished that a user may visit according to the knowledge of the formerly visited pages. WPP problem can be extended and implemented to reduce the access time while surfing the websites. The need to anticipate the needs of the website users to improve accessibility and user engagement is more than apparent now a day. Association rule mining is one of the most significant fields in data mining and knowledge discovery in databases. This paper plans to implement a new web page prediction model using an improved machine learning algorithm. The proposed web page prediction involves three phases (a) Rule Mining, (b) Optimal Rule Selection, and (c) Prediction. Initially, the collected web data is subjected to rule mining process. It is performed using the renowned association rule mining called Apriori algorithm, which is adopted for mining the frequent item set and association rule learning over relational databases. The length of the rule extracted from the Apriori algorithm is long, and it is needed to be reduced for performing the prediction with unique informative rules. Hence, the optimal rule selection is adopted, which uses the hybrid optimization algorithm with the integration of Deer Hunting Optimization Algorithm (DHOA) and Chicken Swarm Optimization (CSO) called Deer Hunting Rooster-based CSO (DR-CSO). Further, the optimally selected rules are subjected to the Machine learning algorithm named Neural Network (NN) for predicting the browsing behavior of the user. Along with the optimal rule extraction, the proposed DR-CSO is used for performing the training in NN. The experimental and comparative results will prove the efficiency of the developed model over existing algorithms.