Plant stress monitoring in IoT systems is typically inferred from environmental variables, providing limited physiological insight. Although abscisic acid (ABA) is a reliable stress indicator, its measurement is destructive and unsuitable for continuous field use. This paper proposes a physiology-aware edge–cloud IoT framework for plant stress classification using non-invasive leaf electrical conductivity integrated with soil and environmental sensing. Leaf electrical conductivity acts as a proxy for stress-related changes in membrane integrity, ionic regulation, and stomatal behavior. A lightweight edge-deployed neural network enables real-time inference, while cloud services support data aggregation and model retraining. The system classifies water, heat, cold, light, and nutrient stress and delivers interpretable outputs via dashboards. Experimental results demonstrate accurate multi-class classification with latency suitable for real-time field deployment, highlighting the benefit of plant-derived physiological signals over environmental proxies.