CNN-Based Emotional Stress Classification using Smart Learning Dataset

Smart learning analytics aims to support researchers investigating mental health by improving the interpretation of the datasets acquired from physiological biomarkers. The key enabler for emotional stress classification are Machine Learning (ML) methods in conjunction with Online Transfer Learning (OTL). The knowledge of high-level characteristics at the top layers is obtained through an optimized Convolutional Neural Network (CNN)-based on emotional stress datasets. Nevertheless, the lack of performance in a real-time environment and the temporal patterns of data acquisition complications and their interpretation motivated us to contribute by tackling these concerns. Therefore, we propose an innovative procedure based on the aforementioned orientation through our research work. Considering mining data streams with concept drifts, we enable the ensemble classifiers. For evaluation, we compare the proposed classification, the LIBrary for Large LINEAR classification (LIBLINEAR) and the Deep Belief Network with Transfer Learning (DBNTL) model. Furthermore, we utilized a multimodal dataset of physical and biological characteristics obtained by fifteen individuals during a lab study. Finally, our framework based on the extracted results has presented more accuracy in classifying an individual's sense of stress. Hence, the proposed method achieves higher efficiency than the state-of-the-art models.