Automatic and Explainable Labeling of Medical Event Logs With Autoencoding

Area of Science:

• Health Informatics
• Data Science
• Machine Learning

Background:

• Process mining is valuable for extracting knowledge from patient pathways.
• Medical event logs are complex, using diverse codes, making event labeling challenging.
• Efficiently labeling these events is crucial for accurate process mining.

Purpose of the Study:

• To present an innovative methodology for handling complex events in medical event logs.
• To develop accurate event labels using clustering in latent space via autoencoding.
• To provide explanations for the created labels through event decoding.

Main Methods:

• Utilized autoencoding to map medical events into a latent space.
• Applied clustering techniques to group similar events in the latent space for labeling.
• Employed decoding to explain the characteristics of the generated labels.

Main Results:

• The method successfully identified hidden clusters in sparse binary synthetic data.
• Accurate explanations for the created labels were generated through decoding.
• A case study on real healthcare data confirmed the method's effectiveness.

Conclusions:

• The proposed autoencoding-based methodology is suitable for complex medical event log analysis.
• It enables accurate labeling and knowledge extraction from patient pathways.
• This approach enhances the application of process mining in healthcare.