IoT- and AI-Based Fertilizer Recommendation for Onion: A Systematic Review

Abstract Imbalanced, non-site-specific fertilizer application undermines the productivity and environmental sustainability of onion (Allium cepa L.), an economically important Indian horticultural crop. Blanket recommendations derived from state-level package-of-practice guidelines cannot capture the spatial and temporal heterogeneity of soil nutrient status, moisture, and micro-climate that exists even within a single field. Low-cost IoT sensor networks, cloud/edge computing, and machine learning (ML) now offer a pathway to real-time, site-specific nutrient management (SSNM). This review synthesizes AI-driven fertilizer recommendation frameworks across the sensing layer (capacitive/resistive moisture sensors, ion-selective NPK probes, optical methods), the connectivity layer (LoRaWAN, NB-IoT, Wi-Fi, MQTT/CoAP), and the intelligence layer (regression, tree-ensemble, and deep-learning models, including CNN-based leaf-deficiency diagnosis), and compares forty-one peer-reviewed studies from 2019–2025. Four gaps emerge: (i) scarce multi-season labeled onion datasets, (ii) weak linkage between prediction models and economically optimal fertilizer doses, (iii) poor interoperability between sensor hardware and recommendation software, and (iv) minimal field-scale validation. We propose a five-layer reference architecture, the AI-Driven Fertilizer Recommendation Framework (AIFRF), combining IoT soil monitoring, a hybrid ensemble-deep-learning engine, explainable AI, and a farmer-facing mobile advisory interface for onion, and outline a research agenda centered on open datasets, low-cost nutrient sensing, federated/transfer learning, and multi-location field trials.