Optimization of machine learning classifier using multispectral data in assessment of Ganodermabasal stem rot (BSR) disease in oil palm plantation

Basal Stem Rot (BSR) disease is a severe issue that affects oil palm cultivation in many regions, especially in Malaysia and Indonesia. Most previous studies have utilized manual assessment methods that consume significant labour costs and time in the plantation. Therefore, this study aimed to assess BSR disease severity levels based on unmanned aerial vehicle (UAV) multispectral data by optimizing the machine learning classifiers through Random Forest (RF), Support Vector Machine (SVM), and Maximum Likelihood (ML) model. The study examines the optimum training sample for the machine learning classifier in BSR disease based on the model’s error performances and to select the best machine learning classifier in BSR disease based on a comparison of individual classes. The results of the study showed that the SVM classifier improved the classification accuracy of healthy category and early infection compared to other models. Furthermore, the RF classifier outperformed SVM and ML in selecting the best model classifier in terms of consistent classification and accurate differentiation based on BSR severity levels.