Machine Learning based Emotion classification in the COVID-19 Real World Worry Dataset

Abstract

COVID-19 pandemic has a dramatic impact on economies and communities all around the world. With social distancing in place and various measures of lockdowns, it becomes significant to understand emotional responses on a great scale. In this paper, a study is presented that determines human emotions during COVID-19 using various machine learning (ML) approaches. To this end, various techniques such as Decision Trees (DT), Support Vector Machines (SVM), k-nearest neighbor (k-NN), Neural Networks (NN) and Naïve Bayes (NB) methods are used in determination of the human emotions. The mentioned techniques are used on a dataset namely Real World Worry dataset (RWWD) that was collected during COVID-19. The dataset, which covers eight emotions on a 9-point scale, grading their anxiety levels about the COVID-19 situation, was collected by using 2500 participants. The performance evaluation of the ML techniques on emotion prediction is carried out by using the accuracy score. Five-fold cross validation technique is also adopted in experiments. The experiment works show that the ML approaches are promising in determining the emotion in COVID-19 RWWD. More specifically, the NN method produced the highest average accuracy scores for both emotion and gender classification where a 75.7% and 72.1% average scores were obtained.