Predictive Modeling of Comorbid Depression and Anxiety Symptoms Among Prospective University Students: A GIS-Based and Machine Learning Study

Prospective university students are highly susceptible to mental health issues such as depression and anxiety. This study investigates the prevalence and risk factors associated with comorbid depression and anxiety suffering, integrating GIS and Machine Learning techniques to comprehensively understand their spatial distribution and predictive factors. Data from 1485 participants were collected via a cross-sectional survey, encompassing socio-demographic details, educational backgrounds, psychological factors, and geographic locations. Statistical analyses utilized SPSS for traditional methods, Python for supervised classification algorithms in machine learning, and R (version 4.3.1) with the 'bangladesh' package for GIS analyses. The study revealed that 29.0% of the participants experienced comorbid depression and anxiety. Key risk factors identified included female gender, urban residency, joint family structures, commerce educational backgrounds, self-coaching for admission, dissatisfaction with mock tests, students higher monthly expenditure, past-year suicidal ideation, and a history of mental health issues. In machine learning, XGBoost SHAP identified past-year suicidal ideation as the most significant predictor, while history of suicide completion in family had minimal impact. The GBM model demonstrated high accuracy, precision, and F1-score metrics, whereas CatBoost excelled in logarithmic loss. GIS mapping highlighted regional disparities, particularly in districts within the Chittagong Hill Tracts and other districts showing elevated prevalence rates. This study highlights the need for targeted interventions and supportive policies to address the prevalent mental health challenges among university entrants, which requires collaborative efforts among educational institutions, healthcare providers, policymakers, and local communities.