<![CDATA[Plant leaf diseases pose a significant threat to agricultural productivity, necessitating accurate and efficient identification systems for timely intervention. This study proposes an approach that leverages deep feature extraction using a pretrained ResNet50 model combined with traditional machine learning algorithms to recognize 38 types of plant leaf diseases. Each image was transformed into a 2048-dimensional feature vector, followed by normalization and dimensionality reduction using Principal Component Analysis (PCA). To mitigate the issue of class imbalance in the dataset, random under-sampling was applied at the feature level to ensure equal representation across all classes. Eleven machine learning models were trained and evaluated using 5-fold cross-validation, with performance assessed through accuracy, precision, recall, F1-score, and ROC AUC score. Among the evaluated models, the Support Vector Machine (SVM) achieved the highest accuracy of 99.63%, followed by Logistic Regression at 97.33%, and LightGBM at 96.25%. These models demonstrated strong generalization capabilities in multiclass settings, while simpler classifiers like AdaBoost and Decision Tree yielded lower performance. A comparative analysis of training and test accuracy further highlighted model robustness and overfitting tendencies. The findings emphasize the potential of combining pretrained convolutional neural networks for feature extraction with conventional classifiers to address complex agricultural classification tasks. Future work may explore the inclusion of healthy leaf samples, alternative CNN architectures, and deployment in real-time diagnostic tools to support precision farming and improve crop health monitoring.]]>
