Identifying Rules for Electroencephalograph (EEG) Emotion Recognition and Classification

In order to identify rules of emotion classification in EEG-based emotion recognition, this paper utilizes rule-based classifier and decision tree algorithm to construct EEG-based emotion classification models for four target emotions, namely happy, sad, angry, and relaxed. Considering different bands of frequency in EEG signals, the band pass IIR filter with Chebyshev type II window was applied to separate the EEG signal into gamma, beta, alpha, and theta bands. Time and frequency domain features extraction methods are presented to seek the relevant features within the EEG signals related to emotional states. For emotion classification, we compared three methods: rules algorithm (RIPPER), decision tree algorithm (J4.8), and SVM. To evaluate our proposed method, a real EEG dataset from DEAP database were used. For the recorded 5-channel EEG signals, the result shows that the RIPPER algorithm yields best performance accuracy of 92.01% in binary classification of sad versus relaxed emotional state. In regards with emotion classification accuracy, rule-based classification model performed better compared to decision tree algorithm. A promising result was obtained from features extracted from beta and gamma bands of EEG signals. From the experiment, the rules classifier model generated 10 rules of emotion classification, while the validation of the rules achieved an average accuracy of 81.64% for relaxed emotion class.