Embedded Machine Learning for Water Micronutrient Detection: Trends, Challenges, and Opportunities

Machine learning (ML) on microcontroller-class hardware offers a transformative pathway for real-time micronutrient sensing in agricultural and environmental monitoring. Traditional sensing methods face challenges related to high costs, delayed responsiveness, and scalability constraints in resource-limited settings. This systematic review adhered to PRISMA 2020 guidelines and evaluated global advancements in ML algorithms and deployment strategies for micronutrient sensing, with a focus on edge-optimized approaches suitable for microcontroller-class hardware. Studies published between 2015 and 2025 were retrieved from Scopus, Web of Science, and Google Scholar, with inclusion criteria targeting peer-reviewed, English-language research employing ML techniques for real-time micronutrient sensing on microcontroller-class hardware. Out of 11,713 initial records, 43 studies met all eligibility criteria. A growing adoption of edge-optimized ML frameworks was observed, with statistical modeling reported in 43.9% of studies and 23.5% emphasizing quantized models and latency optimization. Thematic foci included real-time inference (52.38%), algorithm optimization for embedded devices (25.00%), and model benchmarking (11.90%). Notably, 46.5% of studies did not specify model size or latency, limiting reproducibility. ML frameworks deployed on microcontroller-class hardware show substantial potential for enhancing micronutrient tracking and enabling resource-efficient sensing systems. However, technical, infrastructural, and reporting limitations remain significant barriers. Standardized benchmarking protocols, transparent reporting, and cross-disciplinary collaboration are critical for accelerating adoption.