Implementation of Adaptive Genetic Algorithm for classification problems

Genetic algorithms are one of the most commonly used approaches in data mining. In this article, we propose a genetic algorithm approach to the classification problem. Binary coding is used. In this case, the individuals in the population consist of a fixed number of rules representing possible solutions. The evaluation function takes into account her four key factors: error rate, entropy measure, rule consistency, or hole rate. Adaptive asymmetric mutation is applied with a self-adaptive mutation reversal probability of 1-0 (0-1). The generated rules are not disjoint, but they can overlap. The final conclusion of the prediction is based on the voting rules, and the classifier gives equal voting weight to all rules. Based on three databases, we compared several other traditional data mining techniques and approaches, including decision trees, neural networks, and naive Bayesian learning. The results show that our approach outperforms other approaches in both prediction accuracy