The Impact of Parameter Tuning on The Multi-Class Classification of Copra Quality

Abstract

This study aims to classify copra quality into four categories: Overcooked, Partially Dried, Perfectly Dried, and Sunog using the Support Vector Machine (SVM) algorithm. The performance of the SVM model with default parameters was compared to the model after parameter tuning to evaluate its effectiveness in handling multiclass classification.  The results show that parameter tuning improved the model’s performance on several classes. The Partially Dried class achieved the best performance with a precision of 82.0% and a recall of 91.0%, indicating that the model was able to recognize most samples in this class accurately. The Sunog class also showed improvement after tuning, with a precision of 64.5% and a recall of 74.0%.  However, the model still struggled with minority classes. For the Perfectly Dried class, only 2 out of 45 samples were correctly classified. In the Overcooked class, all data were misclassified, resulting in 0% for both precision and recall.  Overall, this study demonstrates that classification success is not only determined by SVM parameter optimization but also influenced by class imbalance and the similarity of features between copra quality classes. Therefore, in addition to parameter tuning, strategies to address data imbalance are needed to achieve more accurate and balanced classification across all classes.