REPA-PRIN

Healthcare infrastructure is a critical societal function (ENISA, 2020) and strategic investment in this area has been planned in Europe (EC, 2020) and Italy (PNRR 2021–27) for the next few years. In this context, healthcare process analytics has been recognised as a vital tool for supporting organisational management, clinical treatment procedures and decision-making processes. Process data generated by healthcare systems includes diverse information such as activity sequences, time spans and resource utilisation. This heterogeneous multimedia data does not follow the standards of healthcare electronic records, meaning it cannot be immediately represented in operational event logs. Healthcare processes may exhibit non-local dependencies between activities or attribute values, requiring data-hungry techniques to explore them. Human-centred processes are shaped by learning curves, where operator or organisational proficiency is a function of accumulated experience. Healthcare process data cannot be investigated by assuming standard and stationary distributions. REPA addresses this issue by overcoming the 'impedance mismatch' between PM and AI-ML. The fine granularity and non-stationarity of observable process data require new techniques for semantic lifting and preparing data at the required level of abstraction, as well as for behaviour analysis and balancing under-represented variants. Multi-view models must be exploited to capture properties from different dimensions. Continuous and interactive learning is required to adapt the models to new and emerging behaviours. Data augmentation and balancing techniques are essential for handling changes to recently specified processes, where historical data cannot be used.