Predicting Anxiety, Depression and Stress in Modern Life using Machine Learning Algorithms

Abstract In the fast-paced modern world, psychological health issues like anxiety, depression and stress have become very common among the masses. In this paper, predictions of anxiety, depression and stress were made using machine learning algorithms. In order to apply these algorithms, data were collected from employed and unemployed individuals across different cultures and communities through the Depression, Anxiety and Stress Scale questionnaire (DASS 21). Anxiety, depression and stress were predicted as occurring on five levels of severity by five different machine learning algorithms – because these are highly accurate, they are particularly suited to predicting psychological problems. After applying the different methods, it was found that classes were imbalanced in the confusion matrix. Thus, the f1 score measure was added, which helped identify the best accuracy model among the five applied algorithms as the Random Forest classifier. Furthermore, the specificity parameter revealed that the algorithms were also especially sensitive to negative results.