ADS Capstone Chronicles Revised

21

accuracies of 0.83. The F1-score of 0.61 illustrated its effectiveness in high-dimensional data. Adaboost: Achieved validation and test accuracies of 0.84. It performed well in ROC-AUC (0.93) and PR-AUC (0.80) metrics, though its MCC (0.64) and F1-score (0.73) showed limitations in handling the minority class. Support Vector Machine with Kernel Trick: Reached validation and test accuracies of 0.84 and 0.85, respectively. The model showcased an F1-score of 0.76, indicating

effective handling of complex decision boundaries. 5.2 Model Performance Findings

The evaluation of machine learning models for detecting potential fraud, waste, and abuse within the healthcare systems reveals insights into their effectiveness, performance, and suitability for the project’s objectives. Key performance metrics, including accuracy, ROC-AUC, PR-AUC, MCC, and F1-score, were analyzed across different models and datasets. The results are detailed in Tables 3 through 5, which display performance across test, validation, and training datasets.

Table 3 Advanced Model Accuracy Scores - Train (Cross-Validation)

145