The detection of such deceptive activities has proven to be a formidable challenge, relying on traditional methods that involve a limited number of auditors manually scrutinizing thousands of claims.
Fraud in Big Medicare Data
Furthermore, the burgeoning prevalence of Medicare fraud schemes has strained the existing investigative resources.
In response to this predicament, researchers from Florida Atlantic University’s College of Engineering and Computer Science have embarked on a groundbreaking study with the objective of pinpointing fraudulent activity within the extensive realm of big Medicare data.
The researchers focused their attention on imbalanced big data and high dimensionality, where the multitude of features complicates calculations.
The study utilized two imbalanced big Medicare datasets, specifically Part B and Part D, covering medical services and prescription drug benefits, respectively.
Employing the List of Excluded Individuals and Entities (LEIE) provided by the United States Office of the Inspector General for labeling, researchers investigated the impact of Random Undersampling (RUS), a robust data sampling technique, and an innovative ensemble supervised feature selection technique.
Random Undersampling (RUS) entails randomly removing samples from the majority class until a specific balance between the minority and majority classes is achieved. The researchers systematically explored various scenarios, combining RUS and supervised feature selection in different configurations, and scrutinized the results.
Read more: Court Dismisses Elon Musk’s $56 Billion Tesla Compensation Plan
Fraud Detection in Big Medicare Data
Published in the Journal of Big Data, the study unveiled that intelligent data reduction techniques significantly enhance the classification of highly imbalanced big Medicare data.
The synergistic application of both RUS and supervised feature selection surpassed models using all available features. Particularly noteworthy was the superior performance achieved by employing feature selection followed by RUS.
The systematic reduction of features through this approach resulted in more understandable models and markedly improved performance compared to using all features.
The study underscored the significance of comprehending the interplay between feature selection and model robustness, offering computational advantages and enhancing the efficacy of fraud detection systems.
Senior author Taghi Khoshgoftaar, Ph.D., underscored the study’s contribution in understanding how models perform classifications with fewer features, facilitating a more reasoned assessment of their effectiveness.
The findings not only confer computational advantages but also hold the potential to significantly enhance the standard of healthcare service by curbing costs associated with fraud. The co-authors of the study include Huanjing Wang, Ph.D., a professor of computer science at Western Kentucky University, and Qianxin Liang, a Ph.D. student in FAU’s Department of Electrical Engineering and Computer Science.
