Coconut Quality Classification Using Random Forest, XGBoost, and Decision Tree Algorithm with Color, Texture, and Shape Feature Extraction

Indonesia is a significant coconut producer, but the quality classification of coconuts is still performed manually, which is labour-intensive, costly, and time-consuming. This study applies the Team Data Science Process to the modelling stage, utilizing machine learning algorithms such as Random Forest, XGBoost, and Decision Tree to develop a coconut quality classification model. The modelling process considers different data split proportions for training, validation, and test sets: 80:10:10, 70:20:10, $60: 20: 20$, and $50: 20: 30$, to identify the optimal proportion for achieving higher accuracy. Feature extraction uses color moments for color features, Gray Level Co-occurrence Matrix (GLCM) for texture features, and Histogram of Oriented Gradients (HOG) for shape features. The dataset comprises $\mathbf{1. 7 5 0}$ images representing five coconut quality classes: standard, broken, rotten, old, and sprouted. The best classification model was achieved using the XGBoost algorithm with an 80:10:10 data split, resulting in an accuracy and recall of $\mathbf{8 2 \%}$ and a precision and F1-score of $\mathbf{8 1 \%}$ on the test data.