Sugarcane Production Forecasting Model of the Northeastern by Artificial Neural Network

Abstract

This paper presents models to predict sugarcane production in the Northeastern of Thailand using back propagation artificial neural network. In order to select appropriate input factors for the model, the researchers studied other related researches about sugarcane production factors that effect to the sugarcane product. There are six factors (province, production year, average rainfall, average temperature, average sugarcane area, and average cane yield data) effect to the sugarcane production. Those factors were considered as the input of an ANN1 model. In order to confirm factors relationships, autocorrelation and regression were used to analyze the factors. The analysis results showed that just five factors (province, production year, average rainfall, average sugarcane area, and average cane yield data) have relationships. These five factors were considered as the input of an ANN2 model. Base on
assumption that major competitive crops in the Northeastern may has effected on the sugarcane production. Therefore, three more factors (sugarcane price, rubber price and cassava price) were considered as the additional factors of the ANN2 models called ANN3. In order to analyze for the appropriate model, those factors data from year 2004 to 2010 were used in the study. The results showed that the best estimate models of the study are as followings: 1) The ANN1 is with architecture of 5:10:1, correlation at 0.9794. 2) The ANN2 model is with architecture of 6:8:10, correlation at 0.9822. 3) The ANN3 model is with architecture of 8:4:1, correlation at 0.9973. The study indicates that the ANN3 with 8:4:1 architecture of neural network has maximum accuracy.