Healthcare insurance fraud detection using data mining.

Area of Science:

• Data Science
• Health Informatics
• Machine Learning

Background:

• Healthcare insurance fraud is a significant challenge, evolving with sophisticated schemes.
• Detecting fraud is complex due to data issues, real-time needs, privacy, and standardization.
• Existing methods struggle with evolving fraud tactics and data complexities.

Purpose of the Study:

• To present a novel fraud detection methodology for healthcare insurance.
• To leverage association rule mining and unsupervised learning for enhanced fraud identification.
• To analyze the effectiveness of the proposed approach on real-world healthcare data.

Main Methods:

• Utilized association rule mining to extract frequent patterns from healthcare transactions.
• Applied unsupervised learning classifiers (IF, CBLOF, ECOD, OCSVM) to identify anomalies.
• Employed the Centres for Medicare and Medicaid Services (CMS) DE-SynPUF dataset (2008-2010).

Main Results:

• Association rule mining combined with unsupervised techniques was faster (868.18s) than baseline anomaly detection (902.24s).
• CBLOF achieved the highest silhouette score (0.114), indicating superior anomaly detection efficacy.
• Descriptive analysis revealed significant relationships among diagnosis, procedure codes, and physicians.

Conclusions:

• The proposed methodology effectively enhances healthcare insurance fraud detection.
• Combining pattern discovery (association rules) with anomaly detection (unsupervised classifiers) improves accuracy.
• This approach offers a robust solution for identifying sophisticated fraudulent activities in healthcare.