Classification rule discovery using learning automata

Abstract

Learning automata (LA) are adaptive decision making units that can learn to choose the optimal action from a set of actions by interaction with an environment (search space). This article proposes the use of the learning automata as a new tool for data mining (namely LA-miner). The basic scheme is utilizing LA as an effective optimizer for searching the rule-set space. In fact, LA-miner searches the rule-set space to discover an effective rule-set which maximizes a predefined fitness function. The fitness function is related to the total true positives, false positives, true negatives, and false negatives. Extensive experimental results on different kinds of benchmarks with nonlinear, overlapping class boundaries and different feature space dimensions are provided to show the powerfulness of the proposed method. The comparative results illustrate that the performance of the proposed LA-miner is comparable to, sometimes better than those of the CN2 (a traditional data mining method) and similar approaches which are designed based on the swarm intelligence algorithms (ant colony optimization and particle swarm optimization) and an evolutionary algorithm (genetic algorithm).