A novel rough set attribute reduction based on ant colony optimisation
P. Ravi Kiran Varma, V. Valli Kumari, S. Srinivas Kumar · International Journal of Intelligent Systems Technologies and Applications · 2015
Rough Sets (RS), the most promising and proven approach for data reduction, has the ability to retain the essence of the data and it does not expect any domain inputs from an expert. RS based reduction however can attain only local minima, so to elaborate the search space it is wise to employ well known and proven artificial intelligence techniques like the ant colony optimisation (ACO). In this work a novel rough set attribute reduction based on ACO, called as NRSACO is proposed, which can identify global optimal attribute set with the help of rough set based mutual information as a heuristic aid for the ants. Few improvements were suggested through which minimum reducts were attained faster and with fewer ants and iterations. Experiments were conducted on 22 UCI datasets, and the results shows that our approach has outperformed in convergence time with comparable or improved classification accuracies.