Thoranin Sujjaviriyasup

DOI: 10.14456/sjst-psu.2021.151

pp. 1160 - 1168

Abstract

A fertilizer marketing or producing sector has played an important role in agricultural productivity and also food
security around the world for many years. However, the demand of fertilizer consumption is uncertain and difficult to be
forecasted by using simple approaches. Therefore, an accuracy of future demand concerning fertilizer is very interesting task to
support decision making. In this research, a hybrid model of feedforward neural networks and differential evolution emphasizing
on architectural evolution is developed to forecast ten datasets of fertilizer consumption and is compared with conventional
models based on five accuracy measures. The empirical results indicated that the developed model can provide more accuracy
than conventional models at 0.05 significance levels. Furthermore, the capability of the developed model can also provide the
highest precision compared with both ARIMA and SVR models. Consequently, the developed model can be a promising tool to
predict future demands of fertilizer consumption.