Exploratory Data Analysis and Algorithmic Comparison for Early Depression Detection

Early depression detection is very significant for timely intervention and treatment in mental health. This study introduced an extensive system for the early detection of depression, utilizing basic experimentation and analysis techniques. The system is built upon the ‘Detailed_PHQ8_Labels.csv’ dataset, derived from the ‘Extended Distress Analysis Interview Corpus (E-DAIC)’. The dataset contains responses to eight key questions from the ‘Patient Health Questionnaire (PHQ-8), covering symptoms such as ‘No Interest’, ‘Depressed’, ‘Sleep’, ‘Tired’, ‘Appetite’, ‘Failure’, 'Concentrating', and ‘Moving’. The proposed methodology used a ‘thresholding technique’ on the PHQ-8 total score to classify depression into four categories: minimal or no depression, early signs of depression, moderate depression, and moderate to severe depression. This study is based on various algorithms of ‘Machine Learning’ and ‘Deep Learning’. Additionally, the study includes advanced technology for multimodal, such as an ‘Ensemble Model (LSRG)’ and late fusion techniques to improve accuracy. This study focused on the ability to combine traditional thresholding methods with advanced ‘Machine Learning’ and ‘Deep Learning’ algorithms for the early detection of depression. It proposed a robust framework for further development in mental health diagnostics in the medical field.