Abstract
The Kingdom of Saudi Arabia is a major producer of date palm fruit, yet publicly available and well-curated datasets that represent local varieties remain limited. We present SODA (Saudi Organic Date fruits dataset), a balanced image dataset of 13 organic Saudi date varieties: Ajwa, Medjool, Khalas, Raziz, Sukari Galaxy, Sukari Rutab, Sukari Mofattal, Shalabi, Shaishi, Safawi, Sagai, Anbara, and Wanan. SODA contains 2,613 images (201 per class) covering whole fruits, pits, and background-removed counterparts captured under controlled conditions. In addition to the raw images, we provide five tabular datasets extracted using pretrained CNN embedders (MobileNetV3, EfficientNetV2, ResNet50, ResNetV2, and InceptionV3). Using Orange Data Mining software, we benchmark six classical classifiers under stratified 10-fold cross-validation. The best configuration—Neural Network with MobileNetV3 features—achieves an AUC of 95.7% (CA 70.9%, F1 71.0%). SODA and the complete Orange workflows are released publicly to support reproducible research, education, and applications such as automated sorting and quality control.