Data Lifting Matters: Evaluating Event Log Construction in Clinical Process Mining

This record contains the Submitted Manuscript (preprint) of the paper "Data Lifting Matters: Evaluating Event Log Construction in Clinical Process Mining", accepted for presentation at the International Conference on Advanced Research in Technologies, Information, Innovation and Sustainability (ARTIIS 2026), Lisbon, Portugal. The paper investigates how alternative data lifting strategies affect event log quality and downstream predictive performance in healthcare process mining. Using COVID-19 clinical pathways from the COVID Data for Shared Learning (CDSL) dataset, the study compares two event log construction configurations: a compressed representation that reduces repeated clinical observations and a non-compressed representation that preserves all extracted events. The experimental evaluation, conducted on 4,479 patient cases, demonstrates that compression reduces the number of generated events from 304,246 to 238,588 and decreases the duplicate event rate from 24.7% to 3.9%, while maintaining comparable performance in ICU admission prediction using Logistic Regression and Random Forest classifiers. The findings highlight the importance of systematically evaluating data lifting decisions, showing that reducing event log redundancy and complexity does not necessarily compromise predictive effectiveness. This document is the original version of the manuscript submitted before peer review. It does not incorporate subsequent revisions, corrections, or Springer copy-editing and typesetting. Funding: This work was supported by the Italian Ministry of University and Research (MUR) under the PRIN 2022 program – Scorrimento delle graduatorie, project code 2022CJWPNA, CUP H53C24000960006, project title: aRtificial intElligence for Process Analytics (REPA).