ADS Capstone Chronicles Revised

24

interface. The final model chosen was the Bagging classifier model. The trained model was then saved as a .pkl file along with the fitted scaler and encoder. An API endpoint using Python was created to receive input data, preprocess it, and return predictions using the trained model. The endpoint used the prediction method of the trained model to generate predictions based on the input features. Afterwards, a simple HTML front-end was developed to allow users to input data and view predictions. The interface communicates with the Flask API endpoint to retrieve predictions and display results. Because the model is hosted on a local instance, additional steps were taken to make it readily available for sharing. Ngrok and Bitly were used for this purpose. Ngrok is a tool that provides secure tunnels to a local host; in doing so it creates a temporary public URL to be shared without being deployed to a public server. Bitly was used to create a shortened URL that was more user-friendly and readable. 6 Discussion Through exploratory data analysis and modeling processes, it was possible to identify overutilization and underutilization of services, which can indicate inefficiencies in resource allocation and areas prone to FWA. Healthcare is an evolving field. To prevent FWA, it is important to allocate resources efficiently and carefully. This project aims to leverage the Market Saturation and Utilization State-County dataset to observe and analyze the saturation of healthcare

providers compared to the beneficiary population across different areas in the region. This study is crucial because it has the potential to identify overutilization and underutilization of services, which may indicate inefficiencies in resource allocation and areas prone to FWA. The results of the analysis enable policymakers and healthcare providers to make informed decisions that improve the delivery of healthcare services and enhance patient outcomes. 6.1 Conclusion This study found that the bagging classifier model surpassed other machine learning models in identifying potential fraud, waste, and abuse (FWA) candidates. The implementation of the model in healthcare analytics could provide practitioners with a more accurate tool for predicting these candidates, thereby enabling better resource allocation and action plans. Non-compliance with FWA regulations can erode patient trust and jeopardize the sustainability of healthcare programs. By pinpointing areas of over-utilization and under-utilization, this model aids in reducing costs, thereby enhancing patient care and minimizing financial losses. 6.2 Recommend Next Steps and Future Studies Future studies should focus on enhancing FWA detection in healthcare services by including more diverse datasets and discovering more machine learning techniques to enhance predictive capabilities. In addition, further refinement of dashboards, applications, and incorporating user feedback will strengthen stakeholder use. To certify that the models

148