A fair and interpretable deep learning approach for healthcare access prediction in underserved communities

Abstract This study develops prediction models for healthcare access and outcomes that leverage deep learning with bias-attenuating modelling approaches across axes of socioeconomic and demographic diversity. It integrates fairness-aware learning techniques, applies data augmentation strategies, and uses hyperparameter optimization to enhance prediction accuracy while minimizing disparities. Moreover, we conduct extensive simulations to assess the trade-offs between model complexity, fairness, and computational efficiency. Our findings show that fairness-aware predictive models are able to significantly reduce prediction bias, often whilst achieving high accuracy for various demographics. The proposed method achieves better fairness and interpretability than conventional models. That study offers critical insights into the potential of AI-enabled health equity solutions and their implications for policy interventions and clinical decision making. Longitudinal studies can further the adaptability and transparency of predictive models for health care. The model maintained consistently high performance across varying levels of healthcare access, with an AUC ranging from 0.94 to 0.99, indicating reduced bias compared to conventional models.