Detecting novel behavior and process improvement with multi-modal process mining

The importance of data has increased, especially with the spread of Internet of Things (IoT)-like technologies as a result of the 4th Industrial Revolution. In order to make sense of the data, valuable sciences have also gained importance in this direction. Data science is one of these sciences. Although it differs from data science in order to make sense of the data, process mining (PM) plays a dominant role in modeling process-based systems by intersecting with data science. When PM intersects with data science and includes other perspective data in the analysis, problems may occur in performing error-free operations due to data science-related errors. One of these processes is; to detect new behavior with the combination of other perspective data (e.g., data perspective). For existing studies, multi-perspective analysis can be performed with the support of data science by including other perspective data in the analysis. Within the scope of this thesis, by adding other perspective data to the control-flow perspective, multi-modal analysis can be carried out with a PM-intensive approach without loss of context. Multi-modal analysis is the analysis of an activity in combination with the attributes that affect that activity. In addition, instead of using ready-made event logs, event logs can be produced through virtual factories that shine with the 4th Industrial Revolution, and improvement can be made on modeled processes without data from real processes. Essentially, with this strategy, the gap for approaches that use other perspective data with loss of context is filled, but unlike existing process improvements, process improvement can be worked on without waiting for real system event logs. With these advantages, a PM framework that can produce event logs between independent units (i.e., from producer to processor) is proposed. The transmission of these logs between units in a distributed structure is carried out with a broker-based architecture. Root-cause analysis (RCA) can be performed with the help of metrics through other perspectives added to the control-flow perspective for the taken records, and the detection of new behavior in desired level can be done with multi-modal analysis, without loss of context, unlike especially multi-perspective analysis with data exploration mode. In order to verify the framework within the scope of the thesis, an event log is produced via the Business Process Modeling Notation (BPMN) model found with the repairExample.xes event log. Verification is carried out for the generated event logs with multi-perspective BPMN and multi-modal process discovery. Event logs that are verified are transferred to the processor part. The records taken in the processor part are first written to disk. Afterwards, one of the new event logs is selected and root cause analysis is performed. Through other perspective data added to the control perspective, root cause analysis is done multi-modal. Token and alignment based replay is used for predictor and response. If necessary, process improvement is provided through model repair. Fitness is the most important metric for improvement. Precision and generalization can provide insight into the system. In time-related mode, analysis can become a hybrid with variant and query support. Keywords: Process mining, software architecture, machine learning, multi-modal analysis, quality in process mining, root cause analysis, process enhancement.