Abstract: Modern insurance industries increasingly rely on data-driven solutions to forecast the likelihood of claims, aiming to cut financial losses and streamline workflows. Traditional claim reviews are often inefficient and subjective, fueling the need for smart, automated prediction tools. This project develops a machine learning-based system to estimate which policyholders are most likely to file claims, leveraging detailed customer, vehicle, and policy data. Models like Logistic Regression, Random Forest, and Gradient Boosting are assessed on a dataset containing more than 58,000 samples and 41 variables. Advanced preprocessing and careful feature selection improve model stability. Results indicate that ensemble models outperform classic techniques in both accuracy and reliability, supporting better fraud detection, risk assessment, and premium calculation for insurers

Downloads: | DOI: 10.17148/IJIREEICE.2025.131126