A Reinforcement Learning Based Method for Feature Selection on the Imbalanced Datasets

Feature Selection is simply defined as the process of identifying a small subset of highly predictive features out of a large set of candidates. However، in the presence of special conditions like facing imbalanced dataset، this is a rather different process. In this study we consider feature selection as a Reinforcement Learning problem and use the well-known method، i. e. Monte Carlo، to traverse the state space and select the best subset of features. Since، this is a pragmatic procedure to learn from the environment، it is highly adaptive to all of the conditions. Specifically، at first we consider the state space as a Markov Decision Process (MDP) and then introduce an optimal graph search to overcome the complexity of the problem of concern. This method initially explores the space of feature sets، and then exploits the accomplished experiments. According to the conditions of imbalanced datasets، the criterions that consider the circumstances of skewed classes are used for evaluation. The effectiveness and efficiency of our work is demonstrated through conducting several experiments compared with the other state-of-the-art methods1.