Performance of Optimized Machine Learning in Classification of Philippine Legume Variety
Wendell M. Castillo, James Bryan Tababa, Mark Joseph Asuncion, Von Ryan Marcelo, Heherson Albano, Isaac Angelo M. Dioses, Vince Lloyd Q. Balisi
Abstract
This paper presents a comparative study on the classification of soybean grain varieties using two classical machine learning algorithms: Support Vector Machine (SVM) and K-Nearest Neighbors (KNN). The classification task focuses on distinguishing between CL1 SOY and PSB SY2 varieties based on morphological features extracted from digital images. Both models were initially trained using default parameters and later optimized using GridSearchCV to enhance classification performance. Experimental results show that the baseline SVM model achieved an overall accuracy of 57%, with notably poor performance on the CL1 SOY class. In contrast, the untuned KNN model performed significantly better, attaining an accuracy of $\mathbf{9 2 \%}$. Upon applying hyperparameter tuning, both models exhibited substantial improvements. The tuned SVM model reached 95% accuracy, while the optimized KNN model achieved 96%, with precision, recall, and F1-scores consistently high across both classes. These findings show that hyperparameter adjustment improves the performance of machine learning classifiers. Furthermore, the results show that, when appropriately configured, KNN slightly outperforms SVM in this classification job. The suggested method provides a viable solution for automated soybean variety recognition, which contributes to accurate crop quality evaluation and decision assistance in precision agriculture.