An improved genetic algorithm to minimal test cost reduction

Attribute reduction is a classical problem in rough sets. Test-cost-sensitive attribute reduction is a generalization of the problem. It aims to find a minimal test cost set which preserves the whole information of the decision system. Some heuristic reduction algorithms have been proposed to deal with it. However, the results are unsatisfactory especially on medium-sized datasets, such as Mushroom. In this paper, we propose a new attribute reduction approach called improved genetic algorithm. The new approach adopts the cross generational elitist selection strategy. It selects better individuals from the previous generation and the current population to produce the current generation. It can ensure better individuals maintained from one generation to the next. In addition, the fitness function is different from the existing ones. The proposed approach is compared with the two other existing genetic algorithms through experiments on four UCI datasets. The experimental results show that the new appoach consistently outperforms the existing ones. Therefore it is more appropriate for medium-sized datasets.