Depression Prediction Using Enhanced Machine Learning Pipeline

Abstract

Depression is a common mental health problem that has a big impact on a person’s mood, conduct, and ability to function in general. Existing studies have problems while dealing with a large number of features and selecting suitable algorithms at different stages of model development. However, current complex systems are very difficult to understand and use by doctors. The objective of this study is to propose suitable feature extraction and classification models to predict depression that are both cost-effective and easy to understand. The study concludes that the most important features were contentment with the surroundings, financial stress, sleeplessness, anxiety, and psychological issues such as conflict, abuse, and feeling inferior. The findings suggest that Boruta with logistic regression (LR) had the best accuracy of 93.3%, which is better than existing methods.