Obtaining Stable Feature Selection with Heuristic Algorithms from Density-Based Feature Groups

Abstract

The complexity of multidimensionality is one of the frequently encountered problems in the high-dimensional data space. The fact that multidimensionality in the data space increases and reaches great numbers brings about the problem that the number of non-informative ones among the features associated with the target class increases along with the dataset complexity. The fact that all features included in the high-dimensional data space are not distinctive or do not contain critical information generally leads to difficulties at the learning stage. At this point, the importance of feature selection emerges. Feature selection is a problem of minimum subset selection from the original feature set for the best accuracy estimation. The neglected subject in the feature selection is ensuring that the instability problem of the selected feature sets is brought to a solution. This problem comes into prominence with the feature, in the process of performing the discovery of information from the high-dimensional data space. In this study, in addition to their reliable classifying accuracy, the fact that the stability of the optimal feature groups obtained by using heuristic approaches was high was aimed.