Using Machine Learning to Diagnose Depression: A Review of Research Trends and Algorithms

Depression is one of the most common mental health illnesses that affects millions of people worldwide. In the recent past, there have been numerous studies on depression and mental health at large. Artificial Intelligence (AI) technologies have increasingly been adopted to solve complex human-related problems including the diagnosis and treatment of different diseases. Researchers have been trying to establish if and how Machine Learning (ML), a branch of AI, can be used to diagnose depression to help solve this complex problem of mental health.

In this paper, we seek to review and establish unique research trends in using ML algorithms to diagnose depression as well as establish the most recommended ML algorithm that has shown high accuracy and hence mostly been recommended by many research articles.

The study consisted of 2 phases where in the first phase, 3 major publishers and journals were considered to establish research trends. The phase considered articles published between 2015 to 2023. In the second phase, a total of 20 journal articles with open access and having been published between 2020 and 2023 were considered to establish the most recommended ML algorithm for solving the problem of diagnosing depression among people.

Results from phase 1 of the analysis indicated a sharp increase in research trends in this area. In phase 2, different research articles showed varied accuracy with Deep Learning (DL) showing a high level of accuracy. 4 out of the 7 studies that recommended the use of DL as the preferred algorithm showed an accuracy level of over 96%. The same trends were observed in related statistical measures i.e., F1-Score and Area Under the ROC Curve (AUC) where DL outperformed the other ML algorithms. In some studies, it achieved a high score of 0.995 in AUC and 0.98 in F1-Score. DL was mostly used when a large amount of data was considered i.e., survey and social media data. However, other algorithms like Support Vector Machine (SVM), XGBoost, and Logistic Regression also performed well.

Generally, ML and DL proved to be an efficient tool that can be used to diagnose depression by using data from different sources i.e., social media, Electronic Medical Records (EMR), surveys, interviews, and wearables among others. Though DL proved to be an efficient tool to predict depression, in other studies, other ML algorithms had a relatively good performance and were recommended. This raises the need to compare all the algorithms on a case-by-case basis before deciding which one to use.