Understanding facial expression of children with autism using facial recognition and deep learning

Abstract

A prominent research subject in recent times has been the use of Facial Expression Recognition (FER) through Machine Learning. Being a common topic, the challenging part is emotion detection from these facial expressions which can itself become a new sector in this eld, but it is yet to be implemented in the dataset concerning Bangladesh. In our research, we have taken pictures of children and teenagers aged between 5 to 20 with Autism Spectrum Disorder (ASD) and inaugurated their emotions from their images. It is challenging for autistic individuals to gather socially and empathically, as we understand, so this subject matter di ers from the identi cation of a common human facial expression. Our goal is to detect a way such that their emotions can be perceived authentically and thus, make it easy for them including other individuals around them to interact socially without any barriers. In our paper, we implemented 3 models and 7 Machine Learning Algorithms along with 1 Feature Extraction technique to gure out the best accuracies for each algorithm. The proposed system has showcased 89% accuracy for Convolutional Neural Network (CNN), 90.97% for VGG16, 89.17% for Inception v3 models. Our system has also shown an accuracy of 72% for K-Nearest Neighbors (KNN) and 55% for KNN along with Principal Component Analysis (PCA), 61% for Support Vector Machine (SVM) and 67% for SVM with PCA, 56% for Logistic Regression and 56.36% for Logistic Regression along with PCA, 52% for Linear Discriminant Analysis (LDA) and 54% for LDA with PCA, 46% for Decision Tree and 56.36% for Decision Tree along with PCA and 46% for Na ve Bayes and 38% for Na ve Bayes along with PCA.