Learning Membership Functions for Mining Fuzzy Association Rules Using Learning Automata

The Transactions in web datasets often consist of quantitative data, suggesting that the fuzzy set theory can be used to represent such data. The time duration of web pages browsed by users is one type of data saved on log files, which can be used as an important factor to analyze the browsing behavior of users. In all existing researches for mining fuzzy association rules in web usage data the number and parameters of membership functions considered for the time parameter are assumed to be constant across all web pages. However, the number and parameters of the membership functions used for each web page are different from other web pages. So to address this challenge, in this paper a reinforcement based optimization approach based on learning automata(LA) called LA-OMF is proposed to automatically extract both the number and parameters of trapezoidal membership functions for fuzzy association rules in web data. Also, a new heuristic was proposed to increase the convergence speed of the proposed method and eliminate inappropriate membership functions. The performance of the proposed approach was evaluated and the results were compared with the results obtained using the fuzzy web mining approach on a real dataset. Experiments on datasets with different sizes confirmed that the proposed LA-OMF by extracting the optimized membership functions increased the average efficiency of the objective function and the fuzzy support compared to the uniform membership functions by 39% and 61%, respectively.