Towards continuous water stress classification in tomato plants via fuzzy Hoeffding trees and in-vivo biosensors

Abstract Water stress and drought have a critical impact on plant growth and health, influencing and compromising agricultural productivity. Tools that can predict water stress in crops through quantifiable indicators provide valuable information and facilitate timely interventions aimed at maintaining and/or restoring optimal growth conditions before visible and difficult-to-recover symptoms appear. This study introduces an explainable Plant Health Monitoring System (PHMS), based on the continuous monitoring of water stress parameters in tomato plants using a novel in-vivo biosensor called "Bioristor". Our system integrates an explainable incremental classifier by design, specifically experimenting with the traditional Hoeffding decision tree and its fuzzy variant. By analyzing data from the Bioristor, the system evaluates plant health and classifies it into two distinct categories. Additionally, it employs an incremental learning approach, allowing the model to adapt and update during the monitoring period to maintain high classification performance. This continuous monitoring ensures the early detection of water stress, enabling prompt corrective actions. We present results based on a real-world dataset, leveraging four features derived from ionic currents within the plant sap, as measured by the Bioristor. The system performance was evaluated in terms of classification accuracy and model complexity, yielding promising outcomes. Moreover, the extracted decision rules offer valuable insights for implementing effective countermeasures to sustain plant health for extended periods.