Machine Learning Approches for Prediction of Mental Health Issues in Adolescents: A Comparative Survey

Abstract

Mental health is recognized as a non-communicable disease that impairs human lives, sometimes beyond recovery. While everyone is at risk of developing a mental illness, adolescents are more prone to it due to various factors like hormonal changes, study pressure, social pressure, etc. If mental health goes ignored at this stage, it can cause serious, even fatal problems later on in life, which not only impacts a family but also the young workforce of a country. Hence, constant efforts are being made for the early detection of mental disorders so they can be treated better. Early prediction of mental health issues is a classic machine learning problem relying on patient history and data. In this survey, we discuss a total of 22 previous research papers based on machine learning algorithms and other statistical analysis tools employed for the said task and compare their efficacy. The research papers are categorized into different mental health disorders such as 1) Methods for predicting Depression and Anxiety 2) Methods for Suidial Prevalence 3) Methods for Predicting Autism Spectrum Disorder (ASD) 4) Methods for Predicting Substance Abuse among adolescents. On the basis of accuracy, the performance of machine learning prediction models was compared. CNN models, Random Forest, and XGBoost generally performed better than other models. There is centralized research in Pakistan on mental health based on machine learning so SPSS and other tools are mostly used for data analysis. The findings suggest that Machine learning algorithms can be effective for classifying and early predicting high-risk factors among adolescents