Approaches for identifying U.S. medicare fraud in provider claims data

Area of Science:

• Health Informatics
• Data Science
• Machine Learning

Background:

• Rising elderly population and chronic diseases increase healthcare costs.
• Healthcare fraud, particularly in U.S. Medicare, drains resources and reduces service quality.
• Accurate fraud detection is crucial for protecting healthcare programs.

Purpose of the Study:

• To improve the detection of U.S. Medicare Part B provider fraud.
• To develop and assess machine learning approaches for identifying fraudulent activities.
• To predict a physician's specialty based on performed procedures to detect anomalies.

Main Methods:

• Employed machine learning and data mining strategies.
• Developed a baseline model comparing Logistic Regression and Multinomial Naive Bayes.
• Tested improvement strategies: specialty grouping, class removal, and class isolation.

Main Results:

• Proposed improvement strategies showed mixed results compared to the baseline Logistic Regression model.
• The study demonstrated the potential effectiveness of enhanced detection methods.
• Analysis focused on predicting physician specialty from procedure data.

Conclusions:

• Machine learning offers a promising avenue for enhancing Medicare fraud detection.
• Further research and refinement of strategies are needed for optimal performance.
• Improved fraud detection safeguards healthcare resources and ensures service accessibility.