An Automated Deep Neural Network (DNN) and Support Vector Machine (SVM) Model for Facial Emotion Classification

Facial emotion recognition (FER) is a main artificial intelligence (AI) technology that decodes and analyses emotional expressions on human faces utilizing computer vision methods. FER identifies and categorizes a wide range of emotions automatically utilizing machine learning (ML) models, particularly deep neural networks (DNN), such as surprise, pleasure, sorrow, rage, and others. Dealing using variances in lighting, locations, and facial expressions is one of the challenges of FER, as is ensuring that the method simplifies adequately to various emotions and populations. This study proposes a simple DNN-SVM automatic method as a potential solution to FER. The method extracts features utilizing a DNN, pre-processes images utilizing certain methodologies, and then classifies them utilizing a Support Vector Machine (SVM). The input images are first pre-processed utilizing face detection, histogram equalization, gamma correction, and scaling methods. Second, the images are processed utilizing custom single DNNs to extract deep features. Finally, SVM classifies the resulting features. The experiment shows that the suggested structure, which uses an ensemble of DL methods, outperforms a single model. In summary, the suggested model will improve the efficacy of FER classification outcomes and solve FER2013 with 98.85% accuracy, while also managing imbalanced datasets and multi-class classification, making it suitable for use in online learning applications.