Machine Learning Approaches for Predicting Willingness to Pay for Shrimp Insurance in Vietnam

Insurance premium prediction is a problem for limited-resource farmers. Econometric methods have generated inaccurate premium forecasts. This article investigates the efficacy of machine learning in predicting insurance premium. Machine learning techniques and survey data on willingness to pay were collected from 534 farmers in Ben Tre, Khanh Hoa, Quang Ninh, and Tra Vinh Provinces, Vietnam. The top-performing models were cubist, random forest, and support vector machines. The cubist model, with the highest R-2 and lowest root mean square error, was the most appropriate to forecast premiums. Quantity harvested, total cost, stocking density, and willingness to participate in an insurance program were the top-ranked predictors of premium. Predicted premium payments varied by province. Partial dependence plots showed the economic relationship between predicted premium levels and selected variables. The model results demonstrate that machine learning is useful in forecasting insurance premium and exhibits promise for improving econometric techniques in premium determination.