Food Production Forecasting with Time Series and Ensemble Modeling Methods

Abstract

Safe and sufficient food production is important to the achievement of the sustainable development goal targeting "zero hunger by the year 2030" agreed by all member states of the United Nations upon the summit meeting in 2015. This research supports such goal by performing insightful analysis over a long duration of global food production spanning from the year 1971 to 2020. Analysis methodology adopts the application of time series forecasting using the ARIMA algorithm with varied parameter values as well as the machine learning modeling methods through six learning algorithms, which are linear regression (MLR), support vector regression (SVR), artificial neural network (ANN), random forest (RF), gradient boosting (GB), and AdaBoost (AB). The algorithms MLR, SVR, ANN are in the category of single modeling method that a single model is enough for predicting future value, whereas RF, GB, AB are ensemble in which a group of models are used cooperatively to predict the output. To observe characteristics of modeling results, the global models trained from food production index of 164 countries are compared against the minor scale of Thailand. For time series forecasting results, we found that ARIMA (p,d,q) model yields the best prediction at a global scale when setting the parameters (p,d,q) to be (1,1,1), but the parameter values (1,1,2) works better for the minor scale of a single country. In the case of machine learning modeling methods, the ensemble of gradient boosting produces the most accurate forecasting result in both global and regional scales.