Business process management (BPM) is a discipline concerned with the continuous improvement of operations in organizations. Instead of improving isolated functions, BPM focuses on cross-functional end-to-end processes. Recently, data-driven BPM technologies based on machine learning (ML) and process mining are evolving rapidly. Because of the increasing adoption of process mining and ML, the focus of BPM shifted from model-driven to data-driven approaches. Data-driven BPM promises evidence-based decision-making on process improvements. However, during the last decades, companies have invested resources into generating process models and documentations. These process models and the knowledge they contain remains unused by data-driven BPM. Therefore, the present thesis sets out to combine data-driven and model-driven BPM using declarative conformance checking. Conformance checking is a type of process mining. It identifies deviations between process models and process data. Whereas conformance checking typically employs imperative modeling languages that define every process path explicitly, this thesis suggests using declarative models. Declarative models constrain process paths instead of defining them. Multiple perspectives of process execution, including time, resources, and context, can be integrated into declarative process models by adding data constraints. Declarative models aim to simplify the modeling of flexible processes. Because declarative conformance checking combines data-driven and model-driven BPM by definition, it appears as a promising approach for this endeavor. To design a multi-perspective declarative conformance checking technique and to show its applicability in different areas of BPM, a design science research (DSR) project was conducted. The overall research project is in line with the three-cycle view on DSR that comprises a relevance, rigor, and design cycle. The relevance cycle of this thesis overviews process mining applications in the industrial sector and develops a technology-specific process mining maturity grid for the manufacturing and logistics domain. Properties of relevant process mining applications arise from the overview and maturity grid. Additionally, the rigor cycle of this thesis provides an overview of existing conformance checking techniques and ML-applications in BPM. Lastly, design principles for developing and evaluating conformance checking techniques and ML applications rigorously are presented. The design cycle provides the design of a multi-perspective conformance checking technique based on dynamic condition response (DCR) graphs as declarative modeling language. Additionally, the conformance checking technique was embedded in four different ensemble artifacts combining data-driven and model-driven BPM. The conformance checking technique is used to filter process instances deviating from a DCR graph from process data. In an adaptive case management system, the conformance checking technique can be adjusted to act as an execution engine for DCR graphs. Instead of filtering out deviating process instances, these can be labeled in the process data. The resulting labels can be used to train a machine learning model that predicts process deviations before they occur. Furthermore, the conformance checking technique is used as a simulation engine to check the predictions of a deep neural network for their adherence to a DCR graph. Subsequently, nonsensical predictions can be avoided. In addition, design principles and frameworks facilitating the use of BPM systems, generating comprehensible process models from process data, and identifying processes that are viable for automation finalize the list of artifacts. This thesis shows that declarative conformance checking can be utilized in different scenarios to combine data-driven and model-driven BPM. Part A of this dissertation summarizes and relates the articles provided in Part B. Part B includes 13 research articles, of which nine have been published in conference proceedings and one has been published in the journal "Expert Systems With Applications".
Sebastian Dunzer (Wed,) studied this question.