Parameter learning using ant colony optimization for minimal time cost reduction

The time cost is an important issue in cost-sensitive learning. In this paper, we study the parameter learning using the ant colony optimization for minimal time cost attribute reduction. The attribute set is represented by a scatter diagram with each vertex corresponding to an attribute. Firstly, each an-t travels using the weight and total time cost of attributes. Secondly, the ant deletes redundant attributes once the positive region constraint is met. Thirdly, the pheromone of attributes that the ant has obtained is updated. After a number of ants finished their tasks, the ant yielding the least time cost is selected and the optimal reduct is constructed. Experimental results on four UCI datasets show that the optimal parameters for the minimal time cost attribute reduction are found.