Comparative Analysis of CNN and Random Forest for Cashew Plant Disease Classification

Abstract

Cashew plants are a strategic commodity in Indonesia and are highly susceptible to various diseases, making fast and accurate identification techniques essential to minimize economic losses. This study evaluates the comparative performance of Convolutional Neural Networks (CNN) and Random Forest in classifying cashew plant disease images. Using an experimental quantitative approach, the study utilizes a dataset of 6,549 images divided into five classes: anthracnose, gummosis, healthy, leaf miner, and red rust. The models were validated using a split sampling technique with a ratio of 80:10:10 for training, validation, and testing, and were evaluated based on F1-score as well as accuracy, precision, and recall metrics. The Random Forest model employs manual feature extraction, including color, texture (Gray-Level Co-occurrence Matrix/GLCM), and shape (Hu Moments), whereas the CNN model uses a custom Sequential architecture with automatic feature extraction. The experimental results show that CNN achieves an accuracy of 85.80%, outperforming Random Forest by approximately 9%, which attains an accuracy of 76.95%. The novelty of this study lies in the integration of high-level texture features into the Random Forest model to evaluate the performance limits of conventional machine learning compared to CNN-based automatic feature extraction. The findings indicate that CNN performs better for this dataset. However, further optimization—particularly in handling natural background variations—is still required for practical deployment.