PREDICTING CARGO INSURANCE CLAIMS USING MACHINE LEARNING – A CASE STUDY OF THAILAND’S BORDER TRADE
The surge in cross-border trade in Thailand has led to a significant increase in shipments, subsequently raising the potential risks involved. As a result, both goods owners and carriers are compelled to explore effective measures to mitigate the impact in case of unforeseen incidents affecting their cargo. This has intensified their focus on obtaining comprehensive cargo insurance coverage. However, one ongoing issue that both policyholders and insurers have is precisely predicting whether a policyholder will submit a claim to determine a reasonable price for purchasing an insurance policy. The objective of this study is to evaluate and compare individual classifiers in order to determine which provides the most accurate predictions for cross-border freight insurance. This study looked at the relative performance of XGBoost, Logistic Regression, Light GBM, Gradient Boost, Catboost, and Random Forest approaches for predicting cargo insurance claims. The dataset comprises data sourced from The Insurance Premium Rating Bureau (IPRB) from 2016 to 2022 with a specific focus on road transportation in Thailand's border trade. The findings strongly indicate that GradientBoost is the superior model for handling cargo insurance claim predicting. It shows the best score in multiple metrics such as logloss, ROC AUC, precision, and accuracy.
Cargo insurance, Risk prediction, Predictive model, Machine learning, XGBoost, GradientBoost