Predicting oil palm yield using a comprehensive agronomy dataset and 17 machine learning and deep learning models

Abstract

The rising global demand for oil palm emphasizes the importance of accurate oil palm yield predictions. This predictive capability is critical for effective crop management, supply chain optimization, and sustainable farming practices. However, the oil palm sector faces challenges in yield projection, stressing a noteworthy gap in the application and evaluation of modern machine learning and deep learning technologies. Our study addressed this gap by systematically evaluating 17 machine and deep learning models in predicting oil palm yield, incorporating various agronomic parameters, e.g., soil composition, climatic conditions, plant age, and farming techniques. This holistic approach enhances the application of machine and deep learning in agriculture. Using the feature selection technique and a maximum depth of 32 and 1000 estimators, the Extra Trees Regressor exhibited positive performance, i.e., MSE = 860.36 and an R2 = 0.65, and stands out among the 17 models evaluated. Our findings also showed that incorporating a comprehensive agronomic dataset is critical to accurate yield prediction. Hence, this model and approach have the potential to be a robust decision-making tool for agronomists and farmers in the oil palm industry, setting the stage for future innovations in sustainable agriculture practices.