A Data-driven Process Recommender Framework.

Area of Science:

• Computer Science
• Artificial Intelligence
• Process Mining

Background:

• Complex knowledge-based processes often lack structured guidance, leading to performance variability.
• Existing process mining techniques may not adequately capture temporal dynamics or concurrent activities.

Purpose of the Study:

• To develop a data-driven recommender system for improving complex knowledge-based process performance.
• To provide step-by-step recommendations based on historical process data and contextual information.

Main Methods:

• A novel similarity metric was developed to cluster process traces, incorporating temporal and concurrent activity information.
• A recommender system was built to select prototype process performances based on user-defined context attributes.
• Prototypes were identified by analyzing common activities and their temporal relationships in historical data.

Main Results:

• The system achieved a high F1 score of 0.77 on real-world medical process data, significantly outperforming the ZeroR baseline (F1 score of 0.37).
• 63.2% of recommended enactments were among the top five nearest neighbors of actual historical enactments.
• The framework was validated as an interactive visual analytic tool for process mining.

Conclusions:

• Data-driven decision support systems are feasible for enhancing complex knowledge-based processes.
• The proposed approach effectively leverages historical performance data and context for accurate process guidance.
• This work demonstrates a practical application of process mining for real-world decision support.