CONTRIBUTION OF PRIOR KNOWLEDGE TO PROBABILISTIC PREDICTION OF FAMINE

Abstract

The contribution of prior knowledge in prediction of change in food crop prices using ordinary linear regression (OLR) and Gaussian process (GP) based on a probabilistic approach in famine predictions was established in this study. Prior information was obtained from previous results and incorporated into a new dataset. For GP, both approaches incorporating weight-space view and function-space view were applied and results compared. The function-space view produced a more suitable model than the weight-space view and OLR. Probabilistic inference showed better famine prediction accuracy than the conventional inference approach. Addition of prior information into the prediction framework improved prediction. It is recommended that in addition to the developed model, further modeling should be carried out to include the effects of variables such as bumper harvest, availability of inexpensive alternative foodstuffs for consumption, imported foodstuffs to remedy famine, effect of neighborhood price, and cross-border trade.