Environment-Based Oil Palm Yield Prediction Using K-Nearest Neighbour Regression

Abstract

Oil palm is a profitable tree crop, producing two types of oil from fresh fruit bunch (FFB). Oil palm yield prediction is required for import/ export, global food security, and field management. However, complex variations in oil palm yield on account of weather and soil conditions complicate the predictability. Supervised machine learning models can learn nonlinear patterns from complex agrometeorological data. However, environment-based predictive analysis of oil palm yield with machine learning methods is not widely explored. Therefore, this work presents the application of a non-parametric regression algorithm k-nearest neighbor (KNN) for oil palm yield prediction using weather and soil data. This work utilized 35 years of yield, soil, and weather records from Pahang state Malaysia. Data visualization during preprocessing assessment led to an in-depth understanding of environmental impacts on yield patterns. After model selection and training, the statistical evaluation using six different metrics along with an examination of the model's learning process was performed. Results suggested that a substantial amount of data from multiple sources allows reliable forecasts with machine learning models. It is concluded that machine learning is a great potential tool for oil palm yield prediction as an essential part of precision agriculture.