Hybrid ML and DL Approach for Plant Leaf Disease Detection

Abstract

Plant leaf diseases are a major cause of crop loss and create a serious problem for farmers. Traditional methods and earlier machine learning algorithms are not accurate enough and do not work well in real-time. To overcome these problems, this study develops a hybrid model integrating a neural network–based method with traditional predictive algorithms for accurate plant leaf disease detection. This work considers images from six types of plants apple, corn, grapes, peach, potato and strawberry. The model uses EfficientNetV2S to extract features, and Principal Component Analysis is used to reduce data size, which decreases computational cost and improves model performance. The extracted features are then evaluated using several predictive models such as Support Vector Machine, Random Forest, and XGBoost, where SVM provides strong classification in high-dimensional space, Random Forest improves robustness by reducing overfitting, and XGBoost enhances performance through gradient boosting. Additionally, a stacking ensemble method is used where predictions from SVM, Random Forest, XGBoost and EfficientNetV2S are fused using a meta-learning approach to improve overall predictive accuracy. The proposed framework has been tested on 17,459 images across 19 different plant disease classes. The model gives 94.01% accuracy with consistent results, so it can be used in real time. It also helps farmers make better decisions for plant management.