Feature subset selection using generalized steepest ascent search algorithm

This paper presents a novel generalized steepest ascent algorithm for selecting a subset of features. Our proposed algorithm is an improvement upon the prior steepest ascent algorithm by selecting a better starting search point and performing a more thorough search than the steepest ascent algorithm. For any given criterion function used to evaluate the effectiveness of a selected feature subsets, our method is guaranteed to provide solutions that equal or exceed those of the state-of-the-art sequential forward floating selection algorithm. Experimental results for two real data sets confirm that our algorithm consistently selects better subsets than other well-known suboptimal feature selection algorithms do.