Mining classification rules by using genetic algorithms with non-random initial population and uniform operator

Abstract

Classification is a supervised learning method that induces a classification model from a database and is one of the most commonly applied data mining task. The frequently employed techniques are decision tree or neural network-based classification algorithms. This work presents an efficient genetic algorithm (GA) for classification rule mining technique that discovers comprehensible IF-THEN rules using a generalized uniform population method and a uniform operator inspired from the uniform population method. Initial population is generated by methodically eliminating the randomness by generalized uniform population method. In the subsequence generations, genetic diversity is ensured and premature convergence is prevented by the uniform operator. From the experimental results, it was observed that, this method handled the problems of GAs in the task of classification and guaranteed to get rid of any local solution and rapidly found comprehensible rules.