Decoupling Driving Factors and High-Precision Prediction of Food Security in Central Asia Based on a Coupled PLS-SEM and PSO-LSSVM Model

Abstract

Grain supply and demand affect regional food security; however, the drivers are often unclear, making precise forecasting and policymaking challenging. This study used Central Asia as a case to integrate Partial Least Squares Structural Equation Modeling (PLS-SEM) with particle swarm optimization least squares support vector machine (PSO-LSSVM) to separately identify the drivers of grain supply and demand and enhance prediction accuracy. We analyzed the interannual variations in the production, import/export volumes, consumption, and inventory of wheat, rice, barley, maize, and other grains in Central Asia (1992–2019). We then decoupled the factors affecting wheat production and consumption using PLS-SEM and made predictions by integrating PLS-SEM with the PSO-LSSVM. The results showed that grain supply and demand across Central Asia, primarily driven by wheat production and consumption, declined and later recovered, with a turning point between 1995 and 1998. Kazakhstan exports 44% of its wheat, whereas other countries heavily depend on imports. In Central Asia, the path coefficients (r) of the wheat area and yield on total production were 0.36 and 0.77, respectively, whereas in Kazakhstan, they were 0.37 and 0.81, respectively. Climate and cultivation factors indirectly affect production through wheat yield, whereas yield and consumption influence production through area. Economic growth increased wheat consumption, whereas urban population growth decreased it. In Kazakhstan, wheat exports reduced consumption (r = −0.23) but boosted the economy (r = 0.33), a pattern that was not observed in Central Asia. The coupling model of PLS-SEM and PSO-LSSVM enhanced the prediction accuracy of wheat yield, reducing the error by 10.21% in Central Asia and 32.8% in Kazakhstan. This study offers a novel approach to decouple the driving factors of grain production and consumption and predicts crop yields in regions with limited data availability.