Grading the Sweetness Level for Watermelon on Random Forest and Support Vector Machine

The paper’s main contribution is a lightweight, non-invasive watermelon sweetness classifier using only two easily computed features — RGB color averages and roundness — avoiding costly acoustic sensors, spectroscopy, or deep-learning overhead. Random Forest (100 trees) outperformed SVM (RBF kernel), achieving 90% accuracy on validation and 91% on 100 unseen images, with feature-importance analysis showing color (especially Red) as the strongest sweetness predictor over shape. The work demonstrates that a simple, interpretable, low-infrastructure-cost model can rival more complex CNN-based approaches (e.g., YOLOv8, ResNet50) for practical agricultural quality-control deployment.