Screening Depression in IT Industry Using Machine Learning

Abstract

Depression is at an all-time high. In some countries like the US, depression hasn’t been this high since the Second World War. This is a call for urgent action – improvements in diagnosis and treatment of depression need to be brought about. Diagnosis is the first step in addressing depression. Our research revealed that current methods are insufficient in this regard, both traditional and electronic approaches. We wanted to introduce a considerably better depression detection system, for the IT industry in particular to start off with. The best electronic method at present is the questionnaire approach, which has achieved an accuracy of around 81%. Other e-methods like face recognition and sentiment analysis are infeasible in terms of accuracy and/or ease of implementation. We used the OSMI 2018 dataset to train and compare 9 machine learning models while using the same questionnaire approach. Random Forest turned out to be the best, with an accuracy of around 96%. We have thus arrived at a highly accurate 15-question quiz that helps determine if a tech employee has depression, based on his condition in the workplace and family. We hope to test the approach on site and expand it from the IT industry to the general public, ideally causing an improvement in mental health in the population and drive humanity towards better mental healthcare.