| Risk assessment in life insurance is a critical task, achieved through manual techniques traditionally and while effective is time-consuming and prone to inconsistencies. This study presents a machine-learning pipeline that streamlines predictive modeling and handles challenges such as imbalanced data and an abundance of features by applying methods that both balance data representation and simplify the dataset. Five different classifiers, including Logistic Regression as a baseline, XGBoost, CatBoost, LightGBM, and TabNet are tuned using grid search method and evaluated through cross-validation with metrics such as Accuracy, AUC, Recall, Precision, F1 score, and Cohen’s Kappa. The results demonstrate that the ensemble method XGBoost achieved superior performance across four out of six key metrics. Accuracy 0.5801, AUC 0.8678, F1-score 0.5070, and Cohen’s Kappa 0.4622, outperforming traditional, deep learning techniques and prior benchmarks, while TabNet exhibited the highest recall 0.5250, and CatBoost the highest precision 0.5592. This study demonstrates the potential of machine-learning techniques to offer a more efficient and reliable alternative to traditional risk estimation methods. |
*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.