Evaluation of Trace Alignment Quality and its Application in Medical Process Mining

Area of Science:

• Process Mining
• Healthcare Informatics
• Data Analysis

Background:

• Trace alignment algorithms are vital in process mining for identifying standard procedures and deviations.
• Current methods for evaluating trace alignment results lack comprehensive assessment, especially reference-free approaches.
• Discrepancies in results from different alignment algorithms highlight the need for robust evaluation metrics.

Purpose of the Study:

• To analyze and compare existing trace alignment evaluation methods, identifying their limitations.
• To introduce improved reference-free evaluation methods for assessing alignment quality globally.
• To propose a novel metric for measuring alignment complexity to aid algorithm optimization.

Main Methods:

• Comparative analysis of existing trace alignment evaluation techniques.
• Development and implementation of enhanced global, reference-free evaluation metrics.
• Introduction of a new metric for quantifying alignment complexity.
• Application and validation of the proposed methods on a trauma resuscitation dataset.

Main Results:

• Identified limitations in existing reference-free trace alignment evaluation methods.
• Demonstrated that global assessment improves overall alignment quality optimization.
• Introduced a novel alignment complexity metric usable as an optimization constraint.
• Provided medical explanations for deviations identified using the new evaluation methods.

Conclusions:

• The proposed global, reference-free evaluation methods offer a more comprehensive assessment of trace alignment quality.
• The alignment complexity metric provides a valuable tool for optimizing process mining algorithms.
• The methods are applicable in healthcare, as shown by the trauma resuscitation dataset analysis, aiding in understanding treatment pathways and deviations.