Evaluation of TEAP and AuBT as ECG's Feature Extraction Toolbox for Emotion Recognition System

Abstract

This project involves the assessment of ECG feature extraction toolboxes for emotion recognition system. The objective of this work is to compare the performance between TEAP and AuBT by measuring the accuracy of the classification using the features extracted by each toolbox. Two publicly available datasets: DREAMER and AuBT dataset are used in this work. Only ECG signals from both datasets are extracted using TEAP and AuBT toolbox. Using support vector machine, the result for DREAMER dataset shows that the features extracted using TEAP provide better accuracy in classifying arousal while for valence, AuBT is better with 65.40% and 65.80% respectively. AuBT dataset result shows that AuBT toolbox performed marginally better than TEAP. Thus, for large dataset like DREAMER, the 13 features extracted using TEAP can be considered. However, for a smaller data size sample like the AuBT dataset, 81 features extracted by AuBT toolbox is found to benefit the classification process. Additionally, the result of arousal and valence of DREAMER also indicates that the type of emotion data may influence the suitability of the extracted features.