On the Comparison of Markov Chains-based Models in Process Mining for Healthcare: A Case Study

Abstract

In the last decade, Process Mining has become a significant field to help healthcare process experts understand and gain relevant insights about the processes they execute. One of the most challenging questions in Process Mining, and particularly in healthcare, typically is: how good are the discovered models? Previous studies have suggested approaches for comparing the (few) available discovery algorithms and measure their quality. However, a general and clear comparison framework is missing, and none of the analyzed algorithms exploits Markov Chains-based Models.

In this paper, we propose and discuss effective ways for assessing both quality and performance of discovered models. This is done by focusing on a case study, where the pMiner tool is used for generating Markov Chains-based models, on a large set of real Clinical Guidelines and workflows.