Normalized Mutual Information Feature Selection for Electroencephalogram Data based on Grassberger Entropy Estimator

Recently, Electroencephalogram (EEG) has become increasingly important in the role of psychiatric diagnosis and emotion recognition. However, many irrelevant features make it difficult to identify patterns accurately. Obtaining valid features from electroencephalogram can improve the classification and generalization performance. In this paper, an improved normalized mutual information feature selection algorithm which is based on Grassberger entropy estimator (G-NMIFS) is proposed for EEG data. We employ the k-Nearest Neighbor (kNN), Support Vector Machine (SVM), and Naive Bayes methods to compare the proposed approach with normalized mutual information feature selection using Naive estimator and Miller-adjust method. Experimental results on two EEG data sets show that the proposed method can select relevant subsets and improve classification performance effectively.