Explainable AI for Mental Health Prediction in Drug-Affected Populations with Dragonfly Algorithm and GAN Oversampling

AI can help identify mental health issues in drug users earlier: The study demonstrates that machine learning models can achieve 94.17% accuracy in predicting multiple mental health conditions among people with substance use disorders, potentially enabling detection in regions where clinical assessment is difficult to access.

Behavioral and lifestyle factors matter more than demographics: Feature analysis revealed that sleep quality, physical health, and emotional regulation are the strongest predictors of mental health outcomes in this population, while demographic characteristics had minimal impact—suggesting interventions should focus on modifiable behaviors rather than demographic profiling.

Explainability is essential for clinical trust: The researchers used SHAP (SHapley Additive exPlanations) analysis to make their AI model's decisions interpretable to healthcare providers, addressing a critical gap since "low interpretability" of previous models reduced their clinical utility and adoption potential.

Addressing data imbalance improves fairness: The study used GAN (Generative Adversarial Network) oversampling to handle the challenge that mental health cases are rarer than non-cases in drug-affected populations, helping prevent the AI model from being biased toward predicting "no mental illness" simply because that outcome is more common in training data.

Technical optimization matters for real-world performance: The combination of advanced feature selection, synthetic data generation, and algorithmic hyperparameter tuning (using the Dragonfly Algorithm) significantly outperformed simpler baseline approaches, demonstrating that responsible AI requires careful technical implementation alongside ethical considerations.