Facial Expression Recognition Using Ensemble Learning of Transfer Learning Models

Facial expressions are very important for community-based interactions and are very often used to perform human emotion-based behavioral analysis. The task of recognizing facial expressions automatically from a face image is very challenging in computer vision applications. Facial expression recognition (FER) is used in a large set of applications like human–computer interaction and health and behavioral sciences. In this paper, the concept of ensemble learning for facial expression recognition is used for the development of stacking-based ensemble and weight average ensemble. These ensemble models use two or more base learner on which the training is performed and then combines in a way to perform better than individual base learners for classification. Three pre-trained base learners ResNet50, VGG16, and InceptionV3 are used. The accuracy obtained by the three base learners is 91%, 94%, and 93%, respectively. The proposed stack-based ensemble model achieved an accuracy of 97% in FER. The obtained result shows that the proposed ensemble learning performs FER better than the individual and other ensemble deep learning classifiers.