Federated I-PALIA: Privacy-By-Design Distributed Process Discovery for Duplicated Activities in Healthcare

Multicentric studies in healthcare offer a significant opportunity to leverage cross-institutional insights, enhancing data cardinality and effectively reducing potential biases. However, ensuring patient privacy and addressing data ownership issues represent substantial challenges. Federated Learning (FL) stands out as a reliable approach, allowing for distributed computation without necessitating the sharing of raw data. In a similar fashion, Federated Process Mining has emerged as a compelling and innovative field that enables Process Mining studies to be conducted across various institutions while maintaining the integrity of patient privacy. Despite its promising applications, there remains a significant gap in the literature regarding Federated Process Discovery techniques. In particular, federated adaptations presented in the literature focus on footprint matrix-based approaches that do not take into account specific healthcare-related process characteristics. In this paper, we present a federated adaptation of the I-PALIA algorithm, a control flow discovery technique able to identify duplicated activities. This feature is critical in the healthcare domain, where repeated activities are frequent and difficult to represent using traditional algorithms. The proposed distributed version of the I-PALIA algorithm aims to address these shortcomings, enabling distributed computations in multicentric clinical studies in a privacy-by-design manner.