A Hybrid Deep Learning Model for Detection and Recognition Emotion in a Learning Environment
Facial Emotion Recognition (FER) has significant potential for supporting adaptive learning by providing insights into learners’ emotional states and enabling timely educational interventions. However, many existing FER systems experience challenges related to computational complexity, high-dimensional facial features, and the effective extraction of relevant emotional patterns. This study developed a hybrid deep learning model, termed CNN-LP, which integrates Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, and Principal Component Analysis (PCA) for facial emotion recognition. The CNN component was employed to automatically extract discriminative spatial features from facial images, while the LSTM network was used to learn sequential relationships within the extracted feature representations. PCA was incorporated to reduce feature dimensionality and improve computational efficiency while preserving relevant information for emotion classification. The proposed model was trained and evaluated using the Japanese Female Facial Expression (JAFFE) dataset, comprising 213 facial images representing seven emotion classes: anger, disgust, fear, happiness, neutrality, sadness, and surprise. The performance of the model was assessed using accuracy, precision, recall, and F1-score. The experimental results showed that the CNN-LP model achieved an accuracy of 93.90%, precision of 94.74%, recall of 93.90%, and an F1-score of 93.96%. These results indicate that the integration of CNN, LSTM, and PCA provides effective feature representation and reliable emotion classification.
Download Paper