Autism Spectrum Disorder Detection Using Facial Emotion Recognition Mechanism Based on Novel Hybrid DCNN and Optimized ECOC Technique

Deep learning and the recognition of human emotions attracts the researchers lately. Deep neural networks perform excellence in classifying and identifying the image data mainly. The identification of human emotions using facial features is known as human emotion recognition (HER) and used for a variety of things, such as stress analysis. In this article, we examine how to identify an autistic child’s emotions. The concept of multiclass SVM algorithm called Error Correcting Output Code (ECOC) in combination with CNN model for HER is proposed. Optimization of ECOC algorithm increase the efficiency which is experimented and shown through this research work. FER2013 image dataset is used to train and validate the proposed model. According to the experimental analysis, the proposed model performs more efficiently as compared to the hybrid CNN and ECOC classifier. Moreover, the accuracy of proposed model increased with 4.78% as compared to CNN classifier and 2.96% with ECOC. Hence using the powerful concept of optimization one can improve the accuracy of model.