Comparative Analysis of Efficientnet Variants with SVM and Random Forest Classifiers for Dried Clove Quality Classification

The quality of dried cloves is an important factor that determines the selling price, exportability, and competitiveness of products in the global market. The quality classification process, which is still carried out manually in the post-harvest industry, causes inefficiency and inconsistency in the assessment results. To address this issue, this study proposes an artificial intelligence-based automatic classification system with a feature extraction approach using eight variants of EfficientNet (B0-B7) combined with classic machine learning algorithms, namely Support Vector Machine (SVM) and Random Forest (RF). The dataset used consists of$\mathbf{1, 8 2 6}$images of dried cloves divided into three target classes: Intact, Incomplete, and Mixture, with a division ratio of 70 % for training, 15 % for validation, and 15 % for testing. The preprocessing stages include rescaling, resizing, label encoding, feature extraction, and feature standardization. The results of the experiment show that the combination of EfficientNetB3 + SVM provides the best performance with an accuracy of 95.26 %, precision_macro of 95.30 %, recall_macro of 95.26 %, and$\mathbf{F 1}$-score_macro of$\mathbf{9 5. 2 5 \%}$. These findings prove that the combination of EfficientNetB3 feature extraction with the SVM algorithm is capable of accurately and efficiently identifying the quality of dried cloves. The application of this automatic classification system has the potential to increase post-harvest industry productivity, reduce operational costs, and strengthen the competitiveness and traceability of export product quality in the international market.