Explainable Agentic AI-Driven Machine Learning Framework for Property and General Insurance Risk Assessment

An accurate premium estimate is not only a fundamental part of effective property insurance but also a guarantee of financial stability for the insurers and reasonable pricing for the policyholders. In this paper, an Explainable Agentic AI-driven framework has been presented, which combines ensemble machine learning with explainability to enhance risk modeling. Exploratory data analysis was done to investigate the distributions of claims, correlations of features, and policy status, and then experiments were done on a high-performance computing environment. Several models were considered such as Logistic Regression, Decision Tree, Random Forest, Gradient Boosting. The findings show that ensemble techniques gave the best performance with Gradient Boosting giving the highest accuracy of 96.10, precision of 96.22 and a recall of 96.10, closely followed by Random Forest with 95.98, yet the performance of Decision Tree and Logistic Regression was moderate. Compared to base models, SVM (Support Vector Machine), DNN (Deep Neural Network), XGB (Extreme Gradient Boosting) and KNN (K-Nearest Neighbors) performed worse. The feature importance analysis showed that the behavioral and demographic variables were some of the strongest predictors, which provided the transparency of the model decisions and met the regulatory needs. The presented results identify the efficacy of ensemble methods in the provision of high predictive accuracy, as well as improved interpretability to provide a more robust and reliable insurance environment.