Feature selection plays a critical role in pattern classification. Of the various feature selection methods, the sequential floating search (SFS) method is perhaps the most well-known and widely adopted. This paper proposes a feature selection method combining feature ranking and SFS. The proposed feature ranking approach adopts the new idea of false features to rank features based on their importance, and then applies SFS to features that are less important or of lower rank. This approach overcomes issues with the original SFS and extracts more critical features. In addition, most feature selection methods do not consider the problem of multi-class classification. As a result, these methods have difficulty achieving good performance when dealing with a greater variety of classes. Therefore, this study adopts a one-against-all strategy to address this issue. The proposed approach divides multi-class classification into several binary classifications and adopts feature selection to derive individual feature subsets. This strategy achieves satisfactory performance in experimental simulations.
International Journal of Innovative Computing, Information and Control 8(3)pt.B, pp.2089-2100