Abstract: Agriculture is pivotal to global food security and economic stability, particularly in developing nations like India, where 2.5 billion smallholder farmers confront climate variability, labor shortages, inefficient weed management, and fertilizer overuse—resulting in 20-30% yield losses and environmental harm. Accurate yield forecasting, precise weed control, and optimized fertilizer application enable data-informed decisions on crop selection, irrigation, input management, and risk mitigation, thereby alleviating uncertainty, elevating productivity, and advancing sustainable practices while informing policy. This study presents an integrated Smart Farming Assistant powered by artificial intelligence, encompassing: a convolutional neural network (CNN) for real-time weed detection from drone imagery (93-99% precision); a random forest classifier for NPK fertilizer recommendations derived from soil, meteorological, and crop data (97% accuracy); and an XGBoost regressor—chosen for its robustness in modeling intricate feature interactions across extensive historical and environmental datasets—delivering yield predictions with \( R^2 > 0.92 \). Deployed on edge IoT platforms, this system curtails herbicide application by 70%, enhances resource efficiency, augments yield by 15-20%, and offers scalable utility for resource-limited agricultural contexts.

Keywords: Smart Farming, AI Agriculture, Weed Detection, Fertilizer Recommendation, Yield Prediction, Convolutional Neural Network, XGBoost Regressor, NPK optimization, Edge Computing.

Downloads: | DOI: 10.17148/IARJSET.2026.13226