Feature Normalization Effect in Emotion Classification based on EEG Signals

Abstract

Normalization of data in classification-based problem is a fundamental task where binary or multi classifier systems integrate it as a sub-system. Normalization can be thought as a mapping function that makes a transformation from one space to another space. Different types of normalization methods have been proposed depending on the data content. Recently, researches are carried out on whether this process is really necessary. In this paper, the performances of the different normalization methods for Electroencephalogram (EEG) signal based emotion classification are evaluated. Support vector machine based binary classifier is used in emotion classification. Different kernel functions for support vector machine are also considered. Although the experimental findings may not reveal a significant performance difference between different types of normalization, the normalization process increases classification performance of the emotion recognition, in general.