Students' Learning Engagement Detection Based on Cognition and Emotion Features: Improved Linknet-Squeezenet...

Student engagement serves a key purpose in online educational environments because it determines learning outcomes. The process of detecting student engagement becomes difficult because students show different cognitive abilities and emotional states. This paper proposes an EEG-based student engagement detection framework using cognition and emotion features. At first, the input EEG signal is preprocessed through a modified Wiener filter to remove the noise. Next, cognition as well as emotion features are obtained from the filtered signal. Wavelet-based features are extracted as the cognition features. Extraction of entropy features, Stockwell transform features and common spatial features are considered as the emotion-based features. Here, improved correntropy features are proposed under the emotion features. The extracted features are subjected to the Improved Principal Component Analysis for dimensionality reduction. After that, emotion and cognition change are estimated for the further detection process. Finally, a hybrid detection model is introduced for detecting the students' engagement based on the information extracted, which constitutes an integration of SqueezeNet and the optimized LinkNet variant. The proposed Improved LinkNet and SqueezeNet model achieved a mean performance value of 0.954, outperforming existing methods such as LeNet, DCNN, Bi-LSTM, SqueezeNet, LinkNet, LRM and RRCNN.