Researchers have developed a new method to improve the explainability of deep learning models used in predictive process monitoring. The proposed technique addresses the computational complexity and loss of detail associated with existing feature attribution methods by introducing a control-flow-aware segmentation algorithm. This algorithm partitions event logs into meaningful segments, enabling the calculation of segment-level SHAP explanations to identify which parts of a process trace influence predictions. The method was validated on synthetic data and real-world event logs from a loan application process and a Dutch municipality's administrative tasks. AI
IMPACT Enhances trust and adoption of deep learning models in business process optimization by providing more interpretable insights.
RANK_REASON The cluster describes a new research paper published on arXiv detailing a novel method for deep learning explainability.
Read on Hugging Face Daily Papers →
- arXiv
- Deep Neural Networks
- Feature Attribution
- loan application process
- municipality of the Netherlands
- Predictive Process Monitoring
- Shap
- Hugging Face
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
