Online Conformance Checking on Degraded Data

Business processes are often supported by multiple information systems. The execution of these processes leaves behind digital traces known as event data. This data captures various information about the activities performed within the business process and can be used for various purposes, such as determining the sequence of their execution. Alongside event data, personal data belonging to data subjects, individuals directly or indirectly related to the process, may also be collected and processed during the execution of these activities. We collectively refer to event and personal data as process data. Due to various factors, such as privacy regulations, resource constraints in streams-processing systems, or network issues leading to packet loss, process data may be reduced compared to its state at the time of creation. In event data, reduction may be witnessed in the volume of data, for instance, certain events may be forgotten or missing. For personal data, the information contained there may be reduced, for instance, some fragments of a customer address may not be available. We generically refer to such (partially) reduced data as degraded process data. In the process mining field, conformance checking techniques assess how well the execution of a business process aligns with the intended process, identifying any deviations. Regarding event data, this involves comparing the sequence of executed activities with the relevant process model. As a result, any undesired activities or activities missing according to the relevant process model are highlighted. For personal data, conformance checking techniques can highlight deviations from the perspective of accessing and processing of data by various activities in the process, in violation of the defined privacy policies. Applying conformance checking techniques to degraded event data indiscriminately, i.e., without accounting for the degradation, will intrinsically lead to inaccurate or misleading insights. To address this issue, in this work, we have developed novel approaches that enable degradation-aware conformance checking of event data. These approaches aim to compensate for missing information and mitigate the impact of degradation. We categorize these approaches into two classes: stateless and stateful. Stateless approaches, as the name may suggest, do not require prior information regarding missing data. Instead, these techniques employ prediction techniques to estimate such missing data. One of these stateless approaches takes guidance from the relevant process model, while other leverages machine learning classification model to predict information regarding missing data. In contrast, stateful approaches rely on the availability of contextual information about the missing data, such as orientation in the relevant process model. We also proposed systematic data degradation approaches to enable process data degradation in a controlled manner. These approaches degrade data while preserving its future utility. In case of event data, we consider future conformance checking as the primary utility, ensuring that it is minimally impacted. For personal data, we proposed various data degradation policies which retain only the information necessary for subsequent process activities, degrading any redundant parts. We evaluated the proposed conformance checking related approaches by deploying them in online conformance checking scenarios, using both real-life and synthetic datasets. There, we introduced data degradation using our proposed data degradation approaches. For performance assessment, we used root mean square error (RMSE) in conformance score, F1 score for detecting non-conformance, and memory footprint in comparison to a baseline offline conformance checking technique. The results establish that our approaches can effectively estimate the conformance of the degraded data while minimizing the loss of accuracy. Similarly, we implemented our personal data degradation approach in Camunda, a process orchestration and automation platform, in order to demonstrate its practicality and effectiveness. We periodically degrade sensitive data of data subject in parallel with the execution of process activities, which reduces the risk of data exposure in potential data breaches. The proposed approaches are well-suited for deployment in environments that require compliance with privacy regulations, such as the GDPR, as well as in memory-constrained settings, including event streams processing systems.