Sugarcane Yield Grade Prediction using Random Forest and Gradient Boosting Tree Techniques

This paper presents a machine learning based model for predicting the sugarcane yield grade of an individual plot. The dataset used in this work is obtained from a set of sugarcane plots around a sugar mill in Thailand. The features used in the prediction consist of the plot characteristics (soil type, plot area, groove width, plot yield/ yield grade of the last year), sugarcane characteristics (cane class and type), plot cultivation scheme (water resource type, irrigation method, epidemic control method, fertilizer type/formula) and rain volume. We use two predictive algorithms: (i) random forest classification, and (ii) gradient boosting tree classification. The accuracies of our machine learning based predictive methods are 71.83% and 71.64%, respectively. Meanwhile, the accuracies of two non-machine-learning baselines are 51.52% (using the actual yield of the last year as the prediction) and 65.50% (the target yield of each plot is manually predicted by human expert), respectively. This shows that our work is accurate enough to be applied for decision making of sugar mill operation planning.