How do I update my model? On the resilience of Predictive Process Monitoring models to change

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Mining

Background:

• Traditional Predictive Process Monitoring (PPM) models are static, built on historical data.
• These rigid models struggle with evolving processes and new behaviors in real-world environments.
• Lack of adaptability limits the effectiveness of PPM in dynamic settings.

Purpose of the Study:

• To evaluate adaptive strategies for PPM models.
• To investigate the use of incremental learning for updating predictive models.
• To enhance the flexibility and accuracy of PPM in dynamic environments.

Main Methods:

• Assessed three distinct strategies for periodic model rediscovery or incremental construction.
• Compared the performance (accuracy and time) of newly learned models against original static models.
• Utilized diverse real and synthetic datasets, including those with Concept Drift.

Main Results:

• Incremental learning strategies demonstrated improved model performance.
• Adaptive models showed better accuracy and efficiency in handling process variations.
• The study confirmed the viability of updating PPM models with new data.

Conclusions:

• Incremental learning algorithms offer a promising solution for enhancing Predictive Process Monitoring.
• Adaptive PPM models are crucial for effectively managing evolving business processes.
• The findings support the adoption of dynamic model updating in real-world PPM applications.