Machine learning for early detection of plant viruses: Analyzing post-infection electrical signal patterns

Abstract

Plant viral diseases significantly threaten global agricultural productivity, necessitating rapid and non-invasive early detection methods to mitigate losses and prevent disease spread. This study presents a novel approach for early detection of viral infections in plants by analyzing electrical signal patterns, focusing on Alfalfa Mosaic Virus (AMV) infection in tobacco plants. A custom-designed, portable signal recording system was developed, incorporating multiple filtering stages at different steps to minimize noise in field conditions without the need for a Faraday cage, ensuring high-quality signal acquisition. Electrical signals were recorded from both healthy and AMV-infected plants before visible symptoms appeared. By extracting a comprehensive range of features in the low-frequency range of 0–4 Hz and carefully examining specific changes in the electrical signals post-infection, significant alterations were observed, characterized by weakened signal strength and increased irregularity. Through precise feature selection techniques, and using only three key features—Median, Autoregressive coefficients, and Autocorrelation—machine learning models, including Support Vector Machine, K-Nearest Neighbors, and Random Forest, achieved approximately 97% accuracy in detecting virus-infected plants even before the appearance of visual symptoms. This represents superior performance with minimal input data compared to existing methods. These results demonstrate the potential of electrical signal analysis combined with machine learning as a practical, rapid, non-invasive, and affordable tool for early virus detection in plants that is easy to use by non-specialists. Implementing this approach could significantly improve disease management strategies and contribute to sustainable agricultural practices by enabling timely interventions in real-world agricultural environments.