Fast relevance-redundancy dominance: Feature selection for high dimensional data

Feature selection is used to select a subset of features for model construction. This reduces dimensionality which is important for simplification, efficiency and reducing overfitting. Filter-based methods are the most scalable, rating features by their relevance to the target variable via appropriate statistics. Browne, et al. proposed a filter feature selection method, called Relevance-Redundancy Dominance (RRD) with useful properties (no threshold setting, adaptability to any statistics etc.), but with a poor scalability. In this paper, we present a scalable version of RRD, called Fast Relevance-Redundancy Dominance which holds the same properties as RRD while improving scalability. To show the effectiveness of the proposed approach we have carried out extensive numerical experiments on high dimensional datasets (DNA microarray datasets) which shows that it outperforms state-of-the-art algorithms.