Data reduction techniques for highly imbalanced medicare Big Data

In the domain of Medicare insurance fraud detection, handling imbalanced Big Data and high dimensionality remains a significant challenge. This study assesses the combined efficacy of two data reduction techniques: Random Undersampling (RUS), and a novel ensemble supervised feature selection method. The techniques are applied to optimize Machine Learning models for fraud identification in the classification of highly imbalanced Big Medicare Data. Utilizing two datasets from The Centers for Medicare & Medicaid Services (CMS) labeled by the List of Excluded Individuals/Entities (LEIE), our principal contribution lies in empirically demonstrating that data reduction techniques applied to these datasets significantly improves classification performance. The study employs a systematic experimental design to investigate various scenarios, ranging from using each technique in isolation to employing them in combination. The results indicate that a synergistic application of both techniques outperforms models that utilize all available features and data. Moreover, reduction in the number of features leads to more explainable models. Given the enormous financial implications of Medicare fraud, our findings not only offer computational advantages but also significantly enhance the effectiveness of fraud detection systems, thereby having the potential to improve healthcare services.