A Simple Deep Q-Network based Feature Selection Method

Abstract

In this paper, a feature selection method based on a Deep Q-Network (DQN), a state generator module as well as a feature selection module is proposed. The state generator module generates a state vector based on the selected feature subset and the summary of dataset. On the other hand, the feature selection module selects features based on the Q-values and a predefined threshold. Our experimental results show that our proposed methods exhibit good performance in terms of training time.