Fraud Detection Analysis in Supplementary Health Insurance Using a Deep Neural Network (DNN) Model

Abstract

This study aimed to design and evaluate a deep neural network (DNN) model capable of accurately detecting fraudulent claims in supplementary health insurance by combining advanced data preprocessing, domain-specific feature engineering, and deep learning techniques. An applied research design was used with a real-world dataset of 20,000 insurance claims collected from a supplementary health insurance provider. Data integration was performed using SQL Server to unify multiple relational tables, including policy details, insured demographics, claim transactions, and disease information. Rigorous preprocessing included removal of irrelevant features, correlation analysis to eliminate multicollinearity (threshold >0.9), dimensionality reduction via Principal Component Analysis (PCA), imputation of missing values, outlier detection with the Interquartile Range (IQR) method, and normalization using standard scaling. A deep neural network was implemented with Keras, consisting of one hidden layer with 32 neurons using ReLU activation and a sigmoid-activated output layer for binary classification. The model was trained using the Adam optimizer and binary cross-entropy loss over 30 epochs with a batch size of 32. Hyperparameter optimization was supported by randomized search to identify the most effective architecture. The DNN model achieved exceptional performance in distinguishing fraudulent from legitimate claims. Precision, recall, and F1-score for both fraud and non-fraud classes reached 1.00, and overall accuracy was also 1.00. The receiver operating characteristic (ROC) curve showed an area under the curve (AUC) of 1.00, confirming perfect classification ability on the test dataset. The results demonstrate that combining domain-driven feature engineering with deep neural networks can produce highly accurate fraud detection models for supplementary health insurance. This approach provides a scalable and adaptable foundation for insurers seeking to minimize fraudulent payouts and enhance operational efficiency while setting the stage for integrating explainability and real-time detection in future systems.