Attribute reduction algorithm based on improved information gain rate and ant colony optimization

Abstract

Solving minimal attribute reduction (MAR) in rough set theory is a NP-hard and nonlinear constrained combinatorial optimization problem. Ant colony optimization (ACO), a new intelligent computing method, takes strategies of heuristic search, which is characterized by a distributed and positive feedback and it has the advantage of excellent global optimization ability for handling combinatorial optimization problems. Having considered that the existing information entropy and information gain methods fail to help to select the optimal minimal attribute every time, this paper proposed a novel attribute reduction algorithm based on ACO. Firstly, the algorithm adopts an improved information gain rate as heuristic information. Secondly, each ant solves a problem of minimum attributes reduction and then conduct redundancy test to each selected attribute. What’s more, redundant detection of all non-core attributes in the optimal solution will be perfomed in each generation. The result of the experiment on several datasets from UCI show that the proposed algorithms are more capable of finding the minimum attribute reduction and can faster converge and at the same time they can almost keep the classification accuracy, compared with the traditional attribute reduction based on ACO algorithm.