Discovering High-Quality Decision Rules from Business Process Event Logs
Organisatsiooni nimi
Software Engineering and Information Systems
Kokkuvõte
Business Process Management focuses on the analysis and improvement of business processes. One of the main aspects of a process is its control-flow, i.e., the activities composing the path followed by each process execution. In this context, a decision point is a state in which the execution of the process must advance through one out of multiple paths. For example, whether a loan application is rejected or pre-accepted after an initial analysis. Typically, these decisions are based on process data, e.g., the loan amount or risk factors associated with the client.
Understanding the reasonings behind these decision points is crucial in order to design potential optimizations of its performance. With this aim, research in decision mining focuses on discovering and analyzing the decision rules that model the logic of such decision points. Different approaches have been proposed to discover decision rules from event logs recording past executions of the process. In this context, the quality of the discovered rules is typically evaluated w.r.t. their accuracy and precision. However, while a decision rule may be accurate, its usability highly depends on its simplicity and interpretability [1].
This Master's thesis proposes to address this problem by collecting a set of measures of goodness to evaluate the quality of decision rules w.r.t. aspects such as interpretability and simplicity. The goal of this thesis is to design a comparative benchmark for existing decision-mining algorithms. The student is expected to set up a collection of state-of-the-art decision mining approaches, and design a pipeline that evaluates the quality of the discovered rules.
[1] Wais, B., Rinderle-Ma, S. (2024). Towards a Comprehensive Evaluation of Decision Rules and Decision Mining Algorithms Beyond Accuracy. Proceedings of the 36th International Conference on Advanced Information Systems Engineering (CAiSE 2024). LNCS, vol 14663. Springer, Cham.
Understanding the reasonings behind these decision points is crucial in order to design potential optimizations of its performance. With this aim, research in decision mining focuses on discovering and analyzing the decision rules that model the logic of such decision points. Different approaches have been proposed to discover decision rules from event logs recording past executions of the process. In this context, the quality of the discovered rules is typically evaluated w.r.t. their accuracy and precision. However, while a decision rule may be accurate, its usability highly depends on its simplicity and interpretability [1].
This Master's thesis proposes to address this problem by collecting a set of measures of goodness to evaluate the quality of decision rules w.r.t. aspects such as interpretability and simplicity. The goal of this thesis is to design a comparative benchmark for existing decision-mining algorithms. The student is expected to set up a collection of state-of-the-art decision mining approaches, and design a pipeline that evaluates the quality of the discovered rules.
[1] Wais, B., Rinderle-Ma, S. (2024). Towards a Comprehensive Evaluation of Decision Rules and Decision Mining Algorithms Beyond Accuracy. Proceedings of the 36th International Conference on Advanced Information Systems Engineering (CAiSE 2024). LNCS, vol 14663. Springer, Cham.
Lõputöö kaitsmise aasta
2024-2025
Juhendaja
Marlon Dumas and David Chapela de la Campa
Suhtlemiskeel(ed)
inglise keel
Nõuded kandideerijale
Tase
Magister
Märksõnad
Kandideerimise kontakt
Nimi
David Chapela de la Campa
Tel
E-mail