Machine Learning Models Comparison for Water Stress Detection Based on Stem Electrical Impedance Measurements

Smart agriculture is a promising solution to improve food production and reduce waste of resources. The idea is to adopt electronics and sensors to monitor key parameters of the crops and integrate these data with farmer knowledge. Sensors monitor both the environment and the plant itself, generating a huge amount of data. Data processing is a key aspect of smart agriculture, and machine learning can help to understand the data and extract the needed feature. In this paper, we present a performance comparison of several machine learning models trained to detect the water stress condition of plants. The dataset used for this study includes the stem electrical impedance, a novel parameter directly measured on the plants. The machine learning models are compared based on three different metrics, and the average accuracy is higher than 85%. The effect of removing the stem electrical impedance results in worse performance of the models, indicating its impact in the application.