A Machine Learning Approach to detect Depression and Anxiety using Supervised Learning

Abstract

Depression and anxiety are among the leading causes of substantial disability in developing countries. According to a study of World Health Organization (WHO) South East Region, Bangladesh ranks highest in anxiety disorders with women being affected twice severely as men. Intervening these orders at an early stage would be cheaper and more effective than later treatment, and thus, we have proposed a model that uses a standard psychological assessment and machine learning algorithms to diagnose the different levels of such mental disorders. In our proposed model we have found the usage and effectiveness of the five different types of AI algorithms: Convolutional Neural Network, Support vector machine, Linear discriminant analysis, K Nearest Neighbor Classifier and Linear Regression on the two datasets of anxiety and depression. Comparing the results on the basis of different measurement metrics (accuracy, recall and precision), our model achieves the highest accuracy of 96% for anxiety and 96.8% for depression using the CNN algorithm. Additionally, our analysis shows that among Bangladeshi women of age 18-35, 7.4% suffers from profound levels of anxiety and 15.6% undergoes chronic depression.