A Heuristic Genetic Algorithm of Attribute Reduction
Hong Shi, Jin-Zong Fu · 2006
An attribute reduction method is proposed based on genetic algorithm (GA) with heuristic information. It separates the approximate core attributes from the whole attributes set, then represents the rest of attributes with a group of genetic chromosomes using binary encoding. This improves the local searching ability of GA in the process of global optimizing. Furthermore, the method designs the fitness function that prefers finding shorter approximate reducts to longer real reducts, which increases the classification accuracy on new data. Experiments of reduction and classification with the proposed method are conducted. The results show this method is effective and efficient with regard to classification accuracy, classifier scale and convergence