Digital Transformation and Administration Innovation 4 Serial Number 11 2026 01 01 Fraud Detection Analysis in Supplementary Health Insurance Using the LSTM Model Fraud Detection Analysis in Supplementary Health Insurance Using the LSTM Model 1 9 10.61838/dtai.209 EN Journal Article 2025 06 11 <p>This study aimed to design and evaluate an advanced Long Short-Term Memory (LSTM) deep learning model to accurately detect fraudulent claims in supplementary health insurance by leveraging sequential data patterns and domain-specific engineered features. An applied research design was used to build a robust fraud detection framework. A dataset of 20,000 health insurance claims was obtained from a supplementary insurance provider, containing both legitimate and fraudulent cases. Raw relational tables — including policy, insured individuals, claims, disease registry, and branch information — were merged into a single structured dataset using SQL. Rigorous data preprocessing was performed: irrelevant variables were removed, highly correlated features were eliminated through correlation analysis, dimensionality reduction was applied via Principal Component Analysis (PCA), and extreme outliers were excluded using the interquartile range (IQR) method. All numerical features were standardized, and class imbalance was addressed by weighting fraudulent cases during training. The processed data were reshaped into sequences suitable for LSTM input and divided into training and testing sets. An LSTM model with 32 hidden units and a Sigmoid output layer was trained using the Adam optimizer and binary cross-entropy loss, with performance validated through k-fold cross-validation. The LSTM model achieved outstanding predictive performance, with an overall accuracy of 100% on the test dataset. Both fraudulent and non-fraudulent claims reached perfect precision, recall, and F1-scores. Macro and weighted averages also recorded 1.00 across all metrics, indicating the model’s ability to detect rare fraudulent events without sacrificing specificity. By combining advanced deep learning with systematic data preparation and domain-informed feature engineering, the proposed LSTM framework effectively identified complex fraud patterns in supplementary health insurance. This approach offers a scalable and reliable solution to strengthen fraud risk management and reduce financial losses in the insurance sector.</p> https://journaldtai.com/index.php/jdtai/article/download/209/187