IFS: An Incremental Feature Selection Method to Classify High-Dimensional Data

Feature selection (FS) is the problem of finding the most informative features that lead to optimal classification accuracy. In high-dimensional data classification, FS can save a significant amount of computation time as well as can help improve classification accuracy. An important issue in many applications is handling the situation where new instances arrive dynamically. A traditional approach typically handles this situation by recomputing the whole feature selection process on all instances, including new arrivals, an approach that is computationally very expensive and not feasible in many real-life applications. An incremental approach to feature selection is meant to address this issue. In this paper, we propose an effective feature selection method that incrementally scans the data once and computes credibility scores for the features with respect to the class labels. The effectiveness of the proposed method is evaluated on high-dimensional gene expression datasets using different machine learning classifiers.