Real-Time Predictive Detection of Healthcare Fraud, Waste, and Abuse Using Heterogeneous Machine Learning Models

Background: Healthcare waste and abuse are persistent challenges that significantly increase costs and reduce efficiency in the U.S. system. The study focuses on locating major system inefficiencies and fraud points to build a real-time detection system which combines different machine learning methods with human survey data.

Methods: The study used a mixed-method approach which combined survey responses from 275 healthcare professionals with structured claim information. We were trained four machine learning models which included Random Forest and XGBoost and Support Vector Machine and Artificial Neural Network using normalized and preprocessed datasets. The model evaluation process used four different metrics which included accuracy and precision and recall and F1-score. The analysis used binary logistic regression to determine which variables predict abnormal activity.

Results: The survey results show that redundant testing at 28% and administrative delays at 21% and inflated billing at 30% stand as the main causes of waste and abuse. Machine learning models reached outstanding predictive results through XGBoost which achieved 94.6% accuracy and 0.92 F1-score and Random Forest which achieved 93.2% accuracy. The analysis showed two main predictors in Claim Frequency Ratio (β = 0.61, p = 0.046) and Average Billing Deviation (β = 0.54, p = 0.037) and Provider Experience showed a weak negative relationship.

Conclusion: The study shows that combining survey data with different machine learning models creates an effective system for detecting healthcare waste and abuse in real-time. The method helps organizations operate in a moral way while making their financial transactions more efficient.