Abstract

MotivationRecent studies have shown that the ensemble feature selection approaches are essential for generating robust classifiers. Existing methods for aggregating feature lists from different methods require use of arbitrary thresholds for selecting the top ranked features and do not account for classification accuracy while selecting the optimal set. Here we present a two-stage ensemble feature selection framework for finding the optimal set of features without compromising on classification accuracy.\n\nMethods and ResultsWe present herein optSelect, a multi agent-based stochastic optimization approach for nested ensemble feature selection. Stage one involves function perturbation, where ranked list of features are generated using different methods and stage two involves data perturbation, where feature selection is performed within randomly selected subsets of the training data and the optimal set of features is selected within each set using the optSelect. The agents are assigned to different behavior states and move according to a binary PSO algorithm. A multi-objective fitness function is used to evaluate the classification accuracy of the agents. We evaluate the system performance using the random probe method and using five publicly available microarray datasets. The performance of optSelect is compared with single feature selection techniques and existing aggregation methods. The results show that the optSelect algorithm improves the classification accuracy compared to both individual and existing rank aggregation methods. The algorithm is incorporated into an R package, optSelect.\n\nContactkuppal2@emory.edu