A comparative study of machine learning and deep learning models for apple leaf disease detection
Year: 2026
Type: article
Abstract: Apple is one of the most economically important fruit crops cultivated worldwide, yet its production is severely threatened by foliar diseases such as apple scab, black rot, and cedar apple rust. Early and accurate identification of these diseases from leaf images is essential for minimizing yield loss and enabling targeted pesticide applications. Although deep learning has demonstrated remarkable success in plant disease classification, a systematic and fair comparison between convolutional architectures and traditional machine learning approaches on a common, leakage-free evaluation protocol remains scarce. This study presents a comparative evaluation of five classification models, namely a convolutional neural network, an artificial neural network, a recurrent neural network with long short-term memory units, a support vector machine, and a random forest, for automated detection of four apple leaf conditions using the Plant Village apple leaf dataset. A unified preprocessing pipeline comprising red-green-blue conversion, spatial resizing, min-max normalization, and label encoding was applied to all images. Deep learning models were trained on an 80/10/10 stratified split with early stopping and learning rate scheduling, whereas traditional models used an 80/20 stratified split with standard scaling and principal component analysis for dimensionality reduction. Performance was assessed using accuracy, precision, recall, F1-score, confusion matrices. The convolutional neural network achieved the highest classification accuracy of 98.60% with uniform precision, recall, and F1score values, followed by the recurrent neural network (88.70%), the support vector machine (84.00%), the random forest (80.40%), and the artificial neural network (76.60 percent). These findings confirm that convolutional architectures are the most effective choice for image-based apple leaf disease classification, while traditional models offer a computationally economical alternative when hardware resources are limited.
Author: Harpreet Kaur
Institution: Punjabi University
Language: English
Topic: Smart Agriculture and AI
Subfield: Plant Science
Domain: Life Sciences
SDG: Zero hunger
Open Access status: hybrid
OpenAlex ID: W7210285077
