From Medical Xpress, by Florida Atlantic University:
- Medicare is frequently targeted by fraudulent insurance claims, with the estimated annual fraud amounting to over $100 billion. Traditional methods of detecting fraud, which involve manual inspection of claims by a limited number of auditors, are often insufficient due to the volume and complexity of the data.
- A study conducted by the College of Engineering and Computer Science at Florida Atlantic University explored the use of big data and machine learning models to detect Medicare fraud. However, handling imbalanced big data and high dimensionality, where the number of features is extremely high, presents significant challenges.
- The researchers tested two big Medicare datasets, Part B and Part D, using a method called Random Undersampling (RUS) and a novel ensemble supervised feature selection technique. RUS works by randomly removing samples from the majority class until a specific balance between the minority and majority classes is achieved.
- The results showed that the combined use of RUS and supervised feature selection outperformed models that used all available features and data. The best performance was achieved by performing feature selection, then applying RUS. This approach led to data reduction, more explainable models, and significantly better performance.
- The study’s findings could have substantial implications for Medicare fraud detection, offering computational advantages and enhancing the effectiveness of fraud detection systems. If properly applied, these methods could significantly reduce costs related to fraud and improve the standard of health care service.