Yield Prediction Model for Sugarcane in Yunnan Province Based on Association Rules and Multiple Linear Regression

Abstract

【Objective】A sugarcane yield prediction model was constructed to predict sugarcane yield in the main sugarcane producing areas in Yunnan Province.【Method】The sugarcane planting data and regional meteorological data of five main sugarcane producing areas in Yunnan Province were selected as the research objects, and nine influencing factors of reservoir number, nitrogen fertilizerapplication, phosphorus fertilizer application, potassium fertilizer application, compound fertilizer application, usage of plastic film, sugarcane planting area, average annual temperature and annual precipitation were analyzed with association rule algorithm, and five strong correlation factors of sugarcane yield were obtained as sample characteristics, which were brought into multiple linear regression algorithm to construct yield prediction model.【Result】According to the test set validation results, the accuracy rates of the models using multiple linear regression algorithm to construct yield prediction models in Pu'er, Lincang, Honghe, Wenshan, and Dehong regions were 81.1%, 89.3%, 67.8%, 85.3%, and 73.7%, respectively; the accuracy rates of the models using association rule algorithm and multiple linear regression algorithm to construct yield prediction models in Pu'er, Lincang, Honghe, Wenshan, and Dehong regions were 95.4%, 92.8%, 97.9%, 94%, and 91.4%, respectively. The improved accuracy rates of models by association rule algorithm were 14.3%, 3.5%, 30.1%, 8.7%, and 17.7%, respectively.【Conclusion】The results showed that the association rule algorithm could improve the accuracy of the multiple linear regression yield prediction model, and the model showed good prediction results in all five main sugarcane producing areas in Yunnan Province, providing a new method for sugarcane yield prediction.

Keywords

• sugarcane yield
• association rules
• multiple linear regression
• prediction model
• production prediction
• sample feature screening