Facial emotion classification model using convolutional neural networks: A supervised training study in seven emotional categories

Abstract

This article aims to contribute to the study of classification of human facial emotions using convolutional neural networks, through supervised training on seven affective categories: Happiness, sadness, anger, disgust, fear, neutral and surprise. This is a study with an automatic experimental approach, based on supervised training and evaluation of the model through data partitioning (train/validation/test). The study focused on the development and evaluation of a facial emotion classification model using convolutional neural networks. The population consisted of a data set of grayscale images of faces; each labeled with a facial emotion: Happiness, sadness, anger, disgust, fear, neutral and surprise. A sample of 36,000 images of size 48×48 pixels was selected; previously centered and cropped focusing on the face. The instruments used were specialized machine learning and deep learning libraries (Machine Learning and Deep Learning). The results demonstrated the training of a model based on convolutional neural networks to recognize facial emotions; reaching a training accuracy of 90% and a validation accuracy of 65%, complemented with metrics such as F1-score and confusion matrix in the seven affective categories, showing good results especially for emotions such as "happiness" and "anger".