A comparative analysis of efficacy of machine learning techniques for disease detection in some economically important crops

Abstract

Early diagnosis of plant diseases is essential for reducing crop losses and improving agricultural production. Plant diseases markedly diminish output and food supply, highlighting the critical necessity for effective diagnostic instruments. Leaf analysis is an effective technique for evaluating plant health, providing information on diseases and deficiencies. This research underscores the significance of machine learning (ML) in predicting plant diseases, utilising its capacity to analyse characteristics from leaf photos and categorize plants as healthy or ill.

This study presents 35 prediction models that integrate seven leading machine learning techniques—Support Vector Machines (SVM), Random Forest (RF), Naïve Bayes (NB), Decision Tree (DT), k-Nearest Neighbor (KNN), Logistic Regression (LR) and Multilayer Perceptron (MLP)—with five crop varieties: maize, apple, grapes, tomato, and bell pepper. An equitable crop selection strategy guarantees the inclusion of grains, fruits, and vegetables. The PlantVillage dataset is employed for model training and testing, with performance assessed by measures such as Accuracy, Precision, Recall, F1-Measure, ROC, and AUC.

The Random Forest method demonstrated superior efficacy, with the greatest accuracy (91.8%), F1-Measure (91.4%), and AUC score (96.2%), surpassing SVM, KNN, and MLP. The Friedman test provided statistical support, affirming the trustworthiness of these findings. This study enhances the comprehension of machine learning applications in agriculture and establishes a comprehensive framework for the early diagnosis of plant diseases, therefore contributing to food security and sustainable agricultural practices.