Explainable Artificial Intelligence (XAI) In Insurance Fraud Detection
5th International Conference on Informatics and Software Engineering, IISEC 2026, Ankara, Türkiye, 5 - 06 Şubat 2026, ss.537-542, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/iisec69317.2026.11418497
- Basıldığı Şehir: Ankara
- Basıldığı Ülke: Türkiye
- Sayfa Sayıları: ss.537-542
- Anahtar Kelimeler: explainable artificial intelligence, Insurance fraud, machine learning, SMOTE, XAI
- TED Üniversitesi Adresli: Evet
Özet
One major challenge facing insurance firms is fraudulent claims. These false reports drive up expenses, lead to increased customer payments, and erode reliability across provider-customer relationships. As claim information grows in size and intricacy, older methods relying on fixed rules or human review struggle more each year. Instead of depending only on those outdated systems, this work tests nine different supervised machine learning models. Models include Random Forest alongside Gradient Boosting, Support Vector Machine, K-Nearest Neighbors, Gaussian Naive Bayes, followed by Decision Trees, AdaBoost, Logistic Regression, and XGBoost at the end. Since real-world fraud cases appear far less often than valid ones, special rebalancing steps shape how models learn patterns. Without such adjustments, results could favor majority outcomes unfairly. To make outputs clearer for users, tools that reveal the reasoning behind decisions are applied afterward. The LIME method provides local awareness without considering the model type, and SHAP results indicate how each feature contributes to the decision-making process. Both SHAP and LIME methods clarify why a claim is considered a fraud case, while the others pass through checks. Based on the findings, XGBoost had the best performance with 98.02% accuracy and 98.02% F1-score. Furthermore, XAI results indicated that fault-related and policy-related features are the most impactful features in detecting fraud cases.