Stacked Auto-Encoder Driven Automatic Feature Extraction for Web-Enabled EEG Emotion Recognition

Abstract

EEG emotion recognition is able to provide a scientific solution for emotional health assessment. Feature extraction is the fundamental procedure. Traditionally, the future set is generated by the existing theories or rules, which is not convincing and objective enough. Therefore, this paper proposes a data-driven automatic feature extraction methodology for web-enabled EEG emotion recognition based on 2-hidden-layer stacked auto-encoder. Since the web-enabled framework provides large scale of EEG data, emotion-related EEG features can be extracted directly from the time-domain raw wave, which is different from the typical feature extraction methods based on rules and experiences. With the optimal experimental parameters setting, the proposed method extracts typical time-domain distinguishable features from the EEG raw data and obtains relatively low classification error rate. This paper takes a step further towards automatic feature extraction for web-enabled EEG emotion recognition and make the entire framework more impersonal and convincing.