Electroencephalogram Based Emotion Detection Using Hybrid LongShort Term Memory
European Journal of Molecular & Clinical Medicine,
2020, Volume 7, Issue 8, Pages 2786-2792
AbstractEmotion detection using physiological signals is an upcoming research extending applications in various domains. One important challenge in detection of inner emotion states is a good predictive rate in order to build any application. In our present work, a hybrid Long Short Term Memory (LSTM) algorithm is proposed based on channel fusion approach. Data is acquired by eliciting emotions using eight 3-D Virtual Reality (VR) videos for eight discrete emotion states. On preprocessed data, 8-level decomposition using Discrete Wavelet Transforms(DWT) is performed, wavelet features and time-domain features are extracted and fed to Hybrid LSTM. The hybrid algorithm is performing well for eight discrete emotion states (happy excited, calm, bored, fear, tensed, sad and relax), with an accuracy rate of 80.05% and 93.24 % for 4 states in categorical form (Valence- Arousal scale). Frequency domain features on various bands exhibited a good predictive rate than time domain features.
- Article View: 45
- PDF Download: 94