Scalable Mesh Networking with Machine Learning for Real-Time Crop Yield Prediction in Resource-Constrained Agricultural Environments

This paper presents a custom-developed, ultra-lowpower mesh network of sensor nodes designed to log environmental parameters such as temperature and humidity across agricultural fields. The system employs a self-organizing, energy-efficient communication framework optimized for long-term deployment in resource-constrained environments. Logged data is combined with historical agricultural datasets to train a Random Forest Regressor specifically tailored for crop yield prediction. The model demonstrates high accuracy and robustness, effectively translating environmental trends into actionable forecasts. This integrated approach offers a scalable, low-cost pathway toward data-informed agricultural planning, enabling farmers to better anticipate outcomes and adapt to evolving climate conditions. A Random Forest Regressor trained on both field and historical data achieved an R2 of 0.98 and MAE of 4770.27 Hg/ Ha, outperforming linear and decision tree models. Real-time sensor data from 2025 was used to generate accurate yield predictions, demonstrating the system's viability for scalable, data-driven precision agriculture. © 2025 IEEE.