Insurance Claims Fraud Detection Based on Machine Learning

When people or organizations purposefully mislead insurance companies in order to receive benefits or payments to which they are not legally entitled, insurance claims fraud takes place. This kind of fraud can take many different forms with respect to different insurance policies, such as life, health, vehicle, and property insurance. Healthcare insurance claims fraud poses a significant threat to the integrity of healthcare systems worldwide, resulting in substantial financial losses and compromised patient care. In this study the aim is to use machine learning to detect and prevent fraud in healthcare insurance claims. By using Medicare dataset from Kaggle which contains 4 files including inpatient, outpatient, beneficiary and target data, a dataset was created which is a combination of these datasets. The dataset is used for supervised classification problem which classifies the Healthcare Provider weather fraud or not. The model was trained using Logistic Regression, Random Forest, Decision Tree, Support Vector Machine and XGBoost, also the ensemble learning was used to create a more powerful model by combining all the models together. And the feature selection approach was used to improve models performance and to compare the results. It was determined that logistic regression shows the best performance in terms of accuracy, F1-score and AUC score. The SHapley Additive exPlanations which is a model explainability approach is used to interpret the prediction of logistic regression.