Prediction of household food insecurity in rural China: an application of machine learning methods

Abstract Ensuring the accurate prediction of food insecurity among rural households is an essential prerequisite for the implementation of effective interventions aimed at mitigating the risk of household food insecurity. While machine learning has demonstrated potential in enhancing prediction the accuracy of predictions related to household food insecurity, its application remains relatively limited in predicting household food insecurity in rural China. Based on a dataset comprising 3-day food consumption records from 1080 rural households in China, calorie intake was selected as a key indicator for measuring household food insecurity. This study employed machine learning algorithms, specifically the random forest (RF) and least absolute shrinkage and selection operator regression (LASSO), alongside traditional econometric methods, to predict household food insecurity. Additionally, it compared the predictive performance of these machine learning algorithms against that of traditional econometric approaches. The result indicates that RF methods exhibited the highest performance in prediction accuracy, achieving an accuracy of up to 65.7%, closely followed by the LASSO methods. Moreover, this study identified household income, food market accessibility and availability as the most feature variables for predicting household food insecurity. Overall, this study not only demonstrates the viability of machine learning techniques in predicting household food insecurity but also offers valuable implications for preventing the occurrence of household food insecurity in rural China and other developing regions.