ADS Capstone Chronicles Revised

5

on the use of neural networks for detecting Medicare fraud, highlighting its superiority over traditional machine learning models such as support vector machines, logistic regression, and decision trees. Their study highlighted the effectiveness of neural networks in detecting fraudulent claims, due to their ability to handle high-dimensional data and detect complex fraudulent patterns. Johnson and Khoshgoftaar’s research provided valuable insights into implementing neural networks for Medicare fraud detection. Despite the model’s effectiveness, the authors acknowledged challenges, such as data imbalance, which can introduce bias toward the majority class. To mitigate this, oversampling and undersampling techniques were applied to balance the training set. They also recognized the difficulty in the interpretability of neural networks and recommended incorporating explainability methods like Local Interpretable Model-Agnostic Explanations (LIME) and Shapley Additive Explanations (SHAP) to make the model’s predictions clearer. Amponsah et al. (2022) proposed an innovative approach to target healthcare fraud through the integration of blockchain technology. The authors advocated for a solution to the worldwide issue, which is healthcare fraud. This solution enhances the detection and prevention of claim processing. The method used blockchain’s decentralized ledger and smart contracts to enhance data integrity and transparency. On the other hand, decision trees were used to

detecting fraudulent activities and reduced the costs associated with fraud investigation. This study highlighted the potential to revolutionize fraud detection and prevention in healthcare organizations. Johnson and Khoshgoftaar (2023) presented a data-centric approach to improve healthcare fraud classification performance and reliability using Medicare claims data. Their research included enriched datasets that detailed features from providers, claims, and beneficiaries sourced from CMS. These datasets aimed to increase the predictive power of machine learning models by capturing fraudulent behaviors within Medicare more effectively. A significant contribution to their study was their implementation of k-fold-by-npi cross-validation, which allowed for the mitigation of biases found in traditional methods. Johnson and Khoshgoftaar also highlighted the importance of feature engineering in enhancing model performance, as demonstrated by the substantial improvements in evaluation metrics such as AUC and Geometric Mean (G-Mean). This allows for a deeper analysis of feature interactions and improved model interpretability. Additionally, the study demonstrated significant increases in the true positive rate (TPR) and true negative rate (TNR), highlighting the datasets’ capacity to detect fraud while minimizing false positives. The articles consistently focused on unsupervised methods applied to highly imbalanced datasets. However, even in the most recent artificial intelligence (AI) literature, ongoing research and

generate fraud detection rules. The technique improved the accuracy of

129