Predicting depression among men who have sex with men in Ghana using machine learning algorithms

Area of Science:

• Mental Health Research
• Social Epidemiology
• Computational Psychiatry

Background:

• Men who have sex with men (MSM) in Ghana experience significant mental health challenges, including depression, exacerbated by stigma and discrimination.
• Depression is often underdiagnosed and undertreated within this vulnerable population.
• Understanding the specific psychosocial factors contributing to depression is crucial for effective intervention.

Purpose of the Study:

• To identify key psychosocial predictors of depression among MSM in Ghana.
• To evaluate the effectiveness of various tree-based machine learning models in predicting depression risk.
• To inform the development of targeted mental health strategies for MSM in Ghana.

Main Methods:

• Employed seven tree-based machine learning classifiers (Decision Tree, Random Forest, Gradient Boosting, AdaBoost, XGBoost, LightGBM, CatBoost) on a dataset of 225 MSM.
• Utilized sociodemographic data, perceived stress, social isolation, behavioral risks, and stigma measures.
• Applied data pre-processing techniques including handling missing values, feature standardization, one-hot encoding, and Synthetic Minority Over-Sampling Technique (SMOTE) for class imbalance, with performance evaluated via 5-fold cross-validation.

Main Results:

• Random Forest model demonstrated the highest accuracy in predicting depression among MSM.
• Key predictors identified include external social isolation, perceived stress, and stigma related to same-sex behavior.
• Other significant factors contributing to depression risk involved resilience, stigma from gender non-conformity, and sense of community belonging.

Conclusions:

• Depression in Ghanaian MSM is strongly associated with social isolation, stress, and identity-based stigma.
• Machine learning, particularly ensemble methods, offers a powerful tool for identifying at-risk individuals.
• Culturally sensitive mental health interventions and policies promoting social support are essential to address depression in this population.