Trust-aware Explainable Federated Learning (TEFL) for Privacy-Preserving Medical Diagnosis
2026 International Conference on Advances in Artificial Intelligence and Machine Learning (AAIML), (2026), pp. 168-174
Janice A. Abellana
a
,
Ephraimuel Jose L. Abellana
b
a College of Computer Studies, FEU Institute of Technology, Manila, Philippines
b School of Hospitality Management, Philippine Women’s University Ermita, Manila, Philippines
Abstract: The increasing adoption of artificial intelligence in healthcare is hindered by critical concerns regarding patient data privacy and the lack of model transparency. Federated learning (FL) addresses privacy by enabling collaborative model training without centralized data sharing; however, existing FL approaches largely overlook the explainability and trustworthiness of the resulting models’ key requirements in clinical decision-making. In this paper, we propose a Trust-Aware Explainable Federated Learning (TX-FL) framework for privacy-preserving medical diagnosis. The proposed framework integrates explainable artificial intelligence (XAI) into the federated learning process by treating explanation stability as a first-class objective rather than a post-hoc analysis. To quantitatively assess explainability in distributed settings, we introduce a novel metric, the Federated Explanation Consistency Score (FECS), which measures the stability of feature attributions across clients and communication rounds. Extensive experiments conducted on benchmark medical datasets demonstrate that the proposed TX-FL framework achieves competitive diagnostic performance while significantly improving explanation consistency compared to standard federated and centralized learning approaches. Furthermore, our results reveal critical insights into how federated training impacts model interpretability under non-IID data distributions. This work advances trustworthy healthcare AI by bridging the gap between privacy preservation, explainability, and clinical trust in federated learning systems.