A Study on the Potency of Hybrid Models: Detecting Diseases in Cucumber Leaves with Pre-trained CNNs and SVM

Abstract

This study investigates a hybrid machine learning approach for the classification of cucumber leaf diseases, combining the strengths of pre-trained Convolutional Neural Networks (CNNs) and Support Vector Machines (SVM). Focusing on a publicly available dataset, we evaluated four CNN architectures—VGG16, ResNet50, AlexNet, and EfficientNetV2—for feature extraction, followed by SVM for the final classification task. The performance of each model was assessed using F1 scores and confusion matrices, with ResNet50 yielding the highest F1 score, indicative of a strong balance between precision and recall. VGG16 and AlexNet showed potential for improvement, which could be realized through model tuning or integration into an ensemble framework. EfficientNetV2, despite a slightly lower F1 score, highlighted the importance of further training to enhance class distinction. The research underlines the importance of a hybrid approach, which harnesses deep learning's feature extraction capabilities with SVM's classification strength, to improve the accuracy of disease detection in cucumber plants.