Process mining with real world financial loan applications: Improving inference on incomplete event logs

Area of Science:

• Business Process Management
• Data Science
• Financial Modeling

Background:

• Financial loan application processes generate large event logs.
• Incomplete data in these logs can lead to inaccurate process models.
• Existing models may not account for data deficiencies effectively.

Purpose of the Study:

• To model a real-life financial loan application process.
• To analyze the impact of incomplete event logs on decision models.
• To explore the efficacy of non-classical probability in improving model accuracy.

Main Methods:

• Analysis of a robust event log (262,200 events, 13,087 applications).
• Development of a decision model for the loan application service.
• Comparative study of classical versus non-classical probability for inference error reduction.

Main Results:

• Incomplete event logs drastically decrease precision and yield biased decision models.
• Non-classical probability significantly reduces the error percentage of inferences.
• The study demonstrates a method to mitigate data absence issues in financial process analysis.

Conclusions:

• Accurate financial process modeling requires addressing data incompleteness.
• Non-classical probability offers a superior approach to classical methods for handling uncertainty in event logs.
• The findings provide a more reliable decision model for financial loan applications.