مريم توكلي زانياني

Heuristic-based process discovery is one of the most important and popular categories of process discovery methods. It consists of two main steps: 1) constructing dependency graph 2) determining the semantic of joints/splits present in the constructed dependency graph. So far, all heuristic-based methods have deficiencies in both steps. The most significant deficiency is that the current methods cannot guarantee that they can select the best set of process model elements. Due to the fact that mathematical programming can eliminate existing defects, this thesis presents some integer linear programming models for extracting the dependency graph and determining the type of dependency graph splits and joins. On the other hand, according to the literature, the event logs of some of the most popular application areas of process mining (such as the healthcare area) are characterized by including rare and infrequent activities and behaviors (i.e., noise). Hence discovering suitable dependency graphs from event logs of the mentioned areas requires considering the special characteristics of these areas. Considering the importance of discovering the processes model of highly noisy and variable areas, in this dissertation, two mathematical programming models are presented to discover the dependency graph of 1) domains with low and medium noise and variability 2) domains with high noise and variability (such as healthcare domain). Moreover, this dissertation has presented a mathematical programming model for discovering the type of dependency graph splits and joins. (Therefore, in this dissertation, a total of three mathematical programming models are presented.) Based on the eva‎luations performed on real and synthesized event logs, compared to the most prominent heuristic-based process discovery methods, the proposed mathematical models lead to process models with not only higher precision and fitness but also more simplicity. Especially, the proposed method outperformed the rival methods significantly in dealing with healthcare event logs and event logs with high noise levels.

كشف فرآيندهاي كسب‌و‌كار , برنامه‌ريزي خطي عدد صحيح , الگوريتم‌هاي كشف فرآيند ابتكاري , كشف گراف وابستگي , كشف نوع انشعابات و اتصالات گراف وابستگي , فرآيندكاوي

Discovering Business Process Models , Integer Linear Programming , Heuristic-based Process Discovery algorithms , Dependency Graph Discovery , Extracting the Semantics of the Splits/Joints in Dependency Graphs , Process Mining

Maryam Tavakoli-Zaniani

Mohammad Reza Gholamian