Performance analysis of different machine learning approaches for single modal facial expression detection

Abstract

Facial expression detection plays a pivotal role in the studies of emotion, cognitive processes, and social interaction. This has potential applications in different aspects of everyday life .For Example, real time face detection, sentiment analysis, CCTV violence prediction. In this thesis, we investigate and analyze the performance of different machine learning approaches for single modal type facial expression detection. With this proposed model, it is observed that the feature extraction techniques incorporated in this work performs better in recognizing disparate expressions than feeding unprocessed raw dataset to the networks. Moreover, this study used Japanese Female Facial Expression (JAFFE) to demonstrate the comparative performance of different classical classifiers and neural network-based approaches and how viable they are in the detection of facial expression from single modal information. Hence this kind of models increase the advancement of facial recognition for more future purposes .Therefore, the proposed model proves the feasibility of computer vision based facial expression recognition for practical applications like surveillance and Human Computer Interaction (HCI). In this system, Principal Component Analysis (PCA) and t-distributed Stochastic Neighbor Embedding (t-SNE) have been used to solve the dimensionality reduction and visual representation of the feature components in a 2D feature space. For classification and recognition tasks we used different classification algorithms like K-nearest Neighbor (KNN), Support Vector Machines (SVM), Gaussian Naïve Bayes, Random Forest, Extra Tree, Ensemble machines and vanilla neural networks. To use the total dataset on this algorithm we used 80% training and 20% testing of the total dataset. Finally the best accuracy result was given by Artificial Neural Network (ANN) which was 90.63 % from the proposed model.