Optimizing Mining Association Rules for Artificial Immune System based Classification

摘要

The primary function of a biological immune system is to protect the body from foreign molecules known as antigens. It has great pattern recognition capability that may be used to distinguish between foreign cells entering the body (non-self or antigen) and the body cells (self). Immune systems have many characteristics such as uniqueness, autonomous, recognition of foreigners, distributed detection, and noise tolerance . Inspired by biological immune systems, Artificial Immune Systems have emerged during the last decade. They are incited by many researchers to design and build immune-based models for a variety of application domains. Artificial immune systems can be defined as a computational paradigm that is inspired by theoretical immunology, observed immune functions, principles and mechanisms. Association rule mining is one of the most important and well researched techniques of data mining. The goal of association rules is to extract interesting correlations, frequent patterns, associations or casual structures among sets of items in the transaction databases or other data repositories. Association rules are widely used in various areas such as inventory control, telecommunication networks, intelligent decision making, market analysis and risk management etc. Apriori is the most widely used algorithm for mining the association rules. Other popular association rule mining algorithms are frequent pattern (FP) growth, Eclat, dynamic itemset counting (DIC) etc. Associative classification uses association rule mining in the rule discovery process to predict the class labels of the data. This technique has shown great promise over many other classification techniques. Associative classification also integrates the process of rule discovery and classification to build the classifier for the purpose of prediction. The main problem with the associative classification approach is the discovery of high quality association rules in a very large space of candidate rules and incorporating these rules in the classification process. The rule search process is also computationally expensive for the small support threshold values which plays very important role in building an accurate classifier.The artificial immune system (AIS) uses powerful information capabilities of the immune system such as feature extraction, learning pattern recognition etc. The clonal selection algorithm of artificial immune system uses the population-based search model of evolutionary computation algorithms that have the capability of dealing with a complex search space. The clonal selection algorithm has good features for searching and optimization. In this work, we studied and optimised an artificial immune system based classification system. We evaluated the performance of the AIS based classification system by computing accuracy at different clonal factors and varying number of generations. We used three standard datasets to compute the accuracy. Experimentally, we find that the system gives highest accuracy with clonal factor 0.4.