Detection of Statistically Significant Differences Between Process Variants Through Declarative Rules

Exploring Business Process Deviance with Sequential and Declarative Patterns

“…Our work is also related to papers that present approaches for the supervised discovery of declarative models from positive and negative traces like [23,24,25]. Recently, declarative patterns have been used to characterize different process variants like in the approaches presented in [26,27].…”

Process mining-driven analysis of COVID-19’s impact on vaccination patterns

“…The latter is usually provided in the form of an imperative process model or as a set of declarative process rules [26] , which rather than capturing the full process behaviour may describe clinical guidelines. Process variant analysis techniques [27] , [28] , [4] , [29] allow one to automatically compare two or more sets of healthcare process executions exhibiting different outcomes (or performance) to identify relevant differences between the executions that may have had an impact on the outcome or performance of the healthcare process. These type of techniques are applied to answer questions such as: what were the differences between the healthcare treatments provided by two different hospitals to patients having the same diagnosis?…”

“…Existing process mining techniques for automated process discovery and variant analysis (e.g., [22] , [21] , [27] , [29] ) are not very effective when dealing with unbounded process instances, because by design they would implicitly (and erroneously, in our context) assume the first event of a trace in the input event log to be the start of the process instance, and the last event of a trace to be the end of the process instance. We can, however, identify the most appropriate start and end events given a process instance.…”

“…Differences captured by variant analysis techniques are either at control-flow level (i.e., process behavioural differences in terms of executed activities) or at performance level (i.e., differences in the execution/hand-over times of/between the process activities). From both classes of techniques, we selected three state-of-the-art tools, based on previous studies evaluations [3] , [28] , [29] which are: Fodina [21] , Inductive Miner [23] , Split Miner [22] (for automated process discovery); and process comparator [27] , fingerprints-based variant analysis [28] , and variant analysis via declarative rules [29] (for variant analysis).…”

“…We ran these six techniques on the Virtual Machine hosting the data, which was equipped with a Xeon(R) CPU E5-4620 v2 @2.60 GHz with 8 virtual processors and 64 GB of RAM. All the techniques have a Java implementation either as a standalone tools [22] , [29] or as a ProM 9 plugins [23] , [21] , [27] , except the fingerprints-based variant analysis [28] whose implementation is on Python. The Java-based tools (including ProM plugins) where provided with 40 GB of heap space.…”