A Hybrid Model to Favor the Selection of High-Quality Features in High Dimensional Domains

Feature selection is a widely recognized challenging task in dealing with application problems with a large number of features and a limited number of training samples. Filters and wrappers are the most popular feature selection strategies, but recent literature shows the emergence of hybrid approaches aiming at combining the strengths of filters and wrappers while avoiding their drawbacks. This paper proposes a new hybrid model for feature selection that takes advantage of a filter method to weight the relevance of each feature. Top-ranked features are selected, in an incremental way, resulting in a set of nested feature spaces of relatively small size. An evolutionary wrapper further refines each space by extracting small subsets of highly predictive features. Extensive experiments on a benchmark microarray dataset state the effectiveness of the proposed approach.