Naïve bayes classification for oil palm leaf disease based on color and texture features

Abstract

This study presents a comparison between standard Naïve Bayes classifier and its Genetic Algorithm-optimized variant for automated classification of oil palm leaf diseases. The system incorporates RGB color features alongside texture features extracted using the Gray Level Co-Occurrence Matrix. A dataset of of 225 JPG images of oil palm leaves, divided into training and testing sets in an 80:20 split is used. The methodology consisted of preprocessing, feature extraction, and classification. In the preprocessing phase, images were manually cropped, resized to 256 × 256 pixels, and background elements were removed. Feature extraction was then performed to obtain RGB color values and GLCM-based texture values, including contrast, correlation, energy, and homogeneity. Classification was conducted using two variants of the Naïve Bayes algorithm: one with default parameters and another optimized via GA for the Laplace smoothing hyperparameter. Model performance was assessed using a confusion matrix, with accuracy, precision, and recall serving as the primary evaluation metrics. Experimental results showed that both models achieved identical performance, with an accuracy of 51%, a precision of 52%, and a recall of 51%. These findings suggest that the Naïve Bayes classifier, even in its baseline form, demonstrates low discriminative performance for oil palm leaf disease detection, and when enhanced through GA-based optimization, it still provides only limited effectiveness. Therefore, this research highlights the need to pursue alternative methodologies, such as deep learning techniques or the adoption of more discriminative feature representations, aimed at improving both the accuracy and robustness of image-based disease detection in agriculture.