Performance Analysis of Machine Learning Algorithms for Classification of Infection Severity Levels on Rubber Leaves

White root disease (WRD) infection in rubber plantations which is caused by Rigidiporus micropores can lead to a significant yield loss. At the early infection stage, it is very difficult to diagnose the disease because infected trees do not exhibit any symptoms. Thus, this study was carried out to investigate the potential application of spectroscopic technology and machine learning algorithms to classify severity level of infected trees at early stage based on spectral data. A total of 50 leaf samples were used in this work, representing five severity levels; healthy, light, moderate, severe and very severe infection. A visible shortwave near-infrared (VSNIR) spectrometer was used to record the spectral data of the leaf samples. The chlorophyll content of each leaf was measured using SPAD meter. Four classification algorithms investigated in this study were artificial neural network (ANN), support vector machine (SVM), knearest neighbour (kNN) and random forest (RF). The result of the study demonstrates good classification accuracy of 90, 82, 78, and 72% for ANN, SVM, kNN and RF, respectively. This work shows that the spectroscopic measurement combined with classification techniques are promising strategy to classify severity level of WRD based on the spectral data of the rubber leaves.