Machine learning based mental health disorder diagnosis

Abstract

In the present society we live in, most people suffer from mental health disorders like depression, anxiety, addictions, loneliness, stress etc. due to various reasons such as complex and competitive lifestyles, personal problems, insecurities, genetic or other illnesses. Even though it’s burning issue among the society, most people do not like to talk about their mental health problems, share them with counsellors or get treatments due to various reasons like high consultation cost, busy lifestyle, ignorance, or even some people are ashamed of the situation, thinking it is a cause of public humiliation or harm for their social status etc. Because of these above-mentioned reasons, people tend to suffer alone or even ended up committing suicide. Some people with the access and exposure of internet tend to search their symptoms and try to find answers on the internet. But people cannot always trust and rely on the information they found on the internet. There is a problem of confidentiality of the data that submitted by the users for these web sites and there are problems of accuracy and reliability of the information that users received by these web sites. There arises a problem of lack of accurate and reliable digital mental health diagnosis platform that people can use prior seeking professional help. According to literature on mental health disorder diagnosis, Mental health disorders can be traced back to simple behavioral symptoms like feeling nervous, tired, stressful, panic, sweating, overreacting, having nightmares, having suicidal thoughts etc. These behavioral symptoms can be easily identified, analyzed, and can be used as a base to determine whether a person need a professional assistance or not. This research is focused on giving user an accurate and reliable information on their mental health status based on a survey consist with above mentioned behavioral measures and symptoms with the help of machine learning and classification techniques such as Linear Multiclass Classification, Neural Networks, Naïve Bayes, Decision Trees and Decision Forest. The main goal of this paper would be to suggest a system which can accurately classify whether the user have normal mental health condition, or whether they belong to any category of mental health disorder such as anxiety, depression, loneliness, or stress based on behavioral measures and finally suggest them further steps needed to be taken. Finally, this research will critically evaluate correlation of these behavioral symptoms with demographical factors such as age, gender, race, education level, residential area of users etc. and cross validate the accuracy and relevance of the trained model within the Sri Lankan context with the help of opinions of mental health experts, psychiatrist within Sri Lanka.