Abstract

Mental health conditions have become a social problem as the number of cases keep rising among the youth on daily basis. It accounts to be a major public health issue, making up for 23% of the burden of disease in the United Kingdom.

Being a multi-factorial condition, it is not easily diagnosed, unlike other somatic diseases where blood samples are used in detecting infections. So, patients and mental health professionals rely on psychological diagnoses, in most cases applying the Diagnostic and Statistical Manual of Mental Disorders (DSM) techniques which is basically asking the patient key designed questions and then confirming a condition using his medical skill and judgment. Between these two parties, there exist multiple questions and responses which can be explored and analysed using Natural Language Processing (NLP). In the quest to find a lasting solution to these medical conditions, an uncleaned and imbalanced dataset generated over a period of five years from a university in the UK was used for this task. The textual dataset was cleaned, pre-processed and used in the training for a predictive model. Corpora of words were generated from the input or answers to the survey questions in the dataset. These were vectorized and transformed as an array and served as an input variable for training the model. The corpora fed in the model was 7260 words and was capped at 2000 columns.

Seven different ML models were utilized and the classification of 11 different mental health conditions was predicted from the dataset. Among the seven models, Support Vector Machine (SVM), outperformed the other models with an accuracy score of 82%. 80% for both Gradient Boosting (GB) and Convolutional Neural Network (CNN), other models like the decision tree classifier and Naive Bayes made 78% and 79% accuracy scores respectively.

Although the aim was not solely to identify the best model, it was subsequently used for developing a predictive system which will serve as a user interface tool for anyone who intends to check his mental state even before making a visit to a professional mental health practitioner.

A Meta-Prediction of Common Mental Health conditions Among University Age Group Using NLP