Abstract
In the vehicle insurance field, many traditional companies use actuarial methods to evaluate the possibility of the insurance claim, whereas more intelligent systems with predictive analytics models are earnestly growing. The article tries to review and analyze the data set provided by a private insurance company and build a predictive scoring model that can reduce possible costs on new customers. The problem is considered as a binary classification problem on a highly imbalanced data set. All required operations were performed for the data set, such as dealing with missing values, feature engineering, dimensionality reduction, and resampling techniques. Then, on the resulted data set, sophisticated and state-of-the-art machine learning algorithms were applied. When dealing with an imbalanced data set, the critical part is a proper choice of resampling techniques and evaluation metrics. As a result, the best performance was achieved by two models, XGBoost and KNN. The percentage of the correctly predicted cases using the recall metric with the best model was more than 92%. The last step, based on the results of the predictive, recommendations were offered to the company on possible improvements in pricing policies.