Hybrid computational intelligent attribute reduction system, based on fuzzy entropy and ant colony optimization

Abstract

Attribute reduction plays a crucial role in reducing the computational complexity and therefore the resource consumptions in the area of artificial intelligence, machine learning and computing applications. Rough sets are a very promising technique in attribute reduction or feature selection. Fuzzy and rough set hybrids have been proven to be more effective in selecting important features from the available data, particularly in the case of real-time data. There is a need for global searching strategies to find the best possible, minimal combination of features, and at the same time to maintain the originality of information. This paper proposes a hybrid computational intelligent attribute reduction system based on fuzzy entropy, fuzzy rough sets, and ant colony optimization, which do not depend on fuzzy dependency degree. Experimentation conducted on several UCI universal benchmark data sets proves this method to be feasible in obtaining minimal feature set with undisturbed or improved classification accuracy when compared to fuzzy entropy and dependency degree-based fuzzy rough quick reduct.