PROCESS-MINING USING RULE AUGMENTED WITH CONDITIONAL RANDOM FIELDS ON SEQUENCES FOR EVALUATING THE RESULTS IN HEALTH CARE EVENT ABSTRACTION

Abstract

Process-mining approaches are geared towards gaining insights into processes by analyzing records of events. Most of the time, the events written down within the event log aren't enough detailed. This causes the software to uncover incomprehensible processes diagrams or processes that do not exactly match the details that are in the log. We demonstrate that when process discovery software can detect an insufficient process model from an event log that is lower level. The general pattern of the event can in certain instances be identified by abstracting the log higher levels of detail. This presents the challenges of creating a bridge between a log that is low-level events and a more detailed view of the log in order to create a more structured and more easily comprehended model of the process is recognized. We have shown that the supervised learning technique can assist in the task of event abstraction for the case that annotations that contain High-level interpretations of low-level events are only available to specific sequences (i.e. trace). We propose a method to create trace features vectors for events dependent on extensions to an event. There is an increasing amount of research about Process-mining in the field of health care, such as oncology, which is the research of cancer. The MIMIC-III dataset contains 16 event tables that can be beneficial in Process-mining. This paper highlights the potential to utilize MIMIC-III to conduct Process-mining in the field of oncology. The findings and the data's quality limitations are reviewed as well as opportunities for further study as well as reflections on the potential of MIMIC-III to facilitate reproducible research on Process-mining.