Intelligent Nano-Fertilizer Management System Using IoT and Machine Learning for Sustainable Development

Intelligent Nano-Fertilizer Management System Using IoT and Machine Learning for Sustainable Development Kiruthika R*1, Senthil Kumar K1, Preetha devi R1, Jyothy Narayanan1, Umamaheswari M S2 1Department of Agricultural Engineering, Nehru Institute of Technology, Coimbatore, TN, India. 2Department of Science and Humanities, Nehru Institute of Technology, Coimbatore, TN, India. Abstract The integration of nano-fertilizers with intelligent sensing and machine learning (ML) technologies offers a promising pathway toward sustainable agriculture. This study proposes an Intelligent Nano-Fertilizer Management System (INFMS) that combines Internet of Things (IoT) based real-time field monitoring with data-driven decision models for efficient fertilizer application. The system employs soil nutrient and environmental sensors connected through an IoT gateway to collect parameters such as soil moisture, temperature, electrical conductivity, pH, and leaf chlorophyll index. The acquired data are processed by supervised ML algorithms (Random Forest, Support Vector Regression, and Artificial Neural Networks) to predict nutrient demand and optimize nano-fertilizer dosage. Experimental validation on trials revealed an improvement in nutrient use efficiency and a reduction in nitrate leaching compared to conventional fertilizer management. The proposed system demonstrates that integrating nano-fertilizers with IoT and ML can significantly enhance resource efficiency, reduce environmental pollution, and promote sustainable agricultural development. Keywords: Nano-fertilizer, IoT, Machine Learning, Precision Agriculture, Sustainable Development, Smart Farming. I. INTRODUCTION Agriculture today faces a dual challenge: ensuring food security for a growing population while minimizing the ecological footprint of farming practices. Excessive use of conventional fertilizers has resulted in soil degradation, groundwater contamination, and greenhouse gas emissions. Traditional fertilizer application methods are based on generalized recommendations rather than real-time soil and crop conditions, leading to poor nutrient-use efficiency often below 50% (Miller et al. 2025). This inefficiency not only increases production costs but also accelerates environmental pollution, threatening long-term agricultural sustainability. To address these issues, the concept of precision agriculture has emerged, emphasizing sitespecific and data-driven management of agricultural inputs. The development of intelligent farming systems that combine smart sensing, predictive modeling, and nanotechnology offers an innovative route to achieve both productivity and sustainability goals (Raj Sondhiya et al 2024). Among these, nano-fertilizers have shown immense potential due to their high reactivity, controlled nutrient release, and capacity to match nutrient availability with plant demand (Dutta et al. 2025). When coupled with the power of the Internet of Things (IoT) and Machine Learning (ML), these materials can be managed intelligently in real time, forming Journal of Dalian University of Technology | ISSN: 1000-8608 Volume 32, Issue 11, 2025| https://jdut.net/ | Page No-307 the basis for the Intelligent Nano-Fertilizer Management System (INFMS) proposed in this study. The present research aims to design and develop an IoT–ML based nano-fertilizer management framework that enables continuous field monitoring, predictive nutrient assessment, and optimized fertilizer delivery. The proposed system seeks to enhance nutrient use efficiency, improve crop yield, and reduce environmental losses while aligning with the United Nations Sustainable Development Goals (SDGs 2, 6, and 12). II. LITERATURE SURVEY a. Nano-Fertilizers for Sustainable Agriculture Nanotechnology has emerged as a transformative tool in agriculture. Nano-fertilizers nutrients engineered at the nanometer scale offer controlled release, higher solubility, and improved bioavailability compared to conventional formulations. According to Chhipa (2019), nano-fertilizers can significantly reduce application frequency and improve nutrientuse efficiency by up to 30%. Raliya and Tarafdar (2013) demonstrated that ZnO nanoparticles enhanced seed germination and root elongation in maize due to improved nutrient uptake and reactive oxygen modulation. Similarly, Rajonee et al. (2020) reported that nano-NPK formulations increased plant growth rate and chlorophyll concentration while minimizing nutrient losses through leaching. Despite these advantages, the large-scale implementation of nano-fertilizers is limited by the absence of intelligent delivery and monitoring mechanisms that ensure their precise application. b. Role of IoT in Smart Farming The Internet of Things (IoT) has revolutionized agricultural monitoring by enabling real-time data acquisition from soil and environmental sensors. IoT systems capture parameters such as soil moisture, pH, temperature, electrical conductivity, and nutrient concentration, which are crucial for adaptive fertilizer management. Studies by Zhao and Li (2022) and Singh et al. (2024) emphasized that IoT-based precision farming systems can improve input use efficiency by providing continuous field intelligence and feedback control. Kumar et al. (2023) developed a low-cost IoT network for real-time nutrient management, which led to a 20% reduction in fertilizer wastage. However, IoT by itself is primarily a data collection tool—it requires intelligent analytics for actionable recommendations. c. Machine Learning in Agricultural Decision Support Machine Learning (ML) offers the analytical capability to interpret complex, nonlinear relationships among environmental, soil, and crop variables. Algorithms such as Random Forest (RF), Support Vector Regression (SVR), and Artificial Neural Networks (ANN) are particularly effective for predictive modeling in precision agriculture. Mohan et al. (2023) reviewed ML applications in nutrient management, highlighting their ability to reduce uncertainty in fertilizer decision-making. Kollu et al. (2023) demonstrated that ML-based nutrient recommendation systems achieved more than 20% higher prediction accuracy than conventional statistical models. Integrating ML with IoT thus transforms data streams into intelligent insights, enabling adaptive management of nano-fertilizer dosage and timing. Journal of Dalian University of Technology | ISSN: 1000-8608 Volume 32, Issue 11, 2025| https://jdut.net/ | Page No-308 d. Research Gap and Objectives Although IoT and ML have been individually applied in agricultural systems, their integration with nano-fertilizer management remains underexplored. Few studies have addressed the dynamic optimization of nano-fertilizer dosing based on real-time field data (Razauddin et al. 2023). This study bridges the gap by proposing an integrated framework on Intelligent Nano-Fertilizer Management System (INFMS) that combines IoT sensing with Machine Learning based decision models to optimize nutrient use in a sustainable, datadriven manner. The specific objectives are: 1) To develop an IoT-based field monitoring system for continuous data acquisition. 2) To design ML models (RF, SVR, ANN) for nutrient demand prediction. 3) To validate system performance through field experiments on nutrient-use efficiency, yield, and nitrate leaching III. MATERIALS AND METHODS a. System Architecture The INFMS architecture comprises three layers: 1) Sensing Layer: Includes soil nutrient sensors (NPK), moisture probes, temperature sensors, pH meters, and leaf chlorophyll sensors connected via NodeMCU and Arduino microcontrollers. 2) Communication Layer: Utilizes LoRa and Wi-Fi modules to transmit real-time field data to a central cloud server. 3) Intelligence Layer: Implements ML models on the cloud platform for nutrient prediction, supported by a web dashboard for visualization and control. Fig. 1 Three-layer architecture of the Intelligent Nano-Fertilizer Management System Journal of Dalian University of Technology | ISSN: 1000-8608 Volume 32, Issue 11, 2025| https://jdut.net/ | Page No-309 The proposed system in fig. 1 shows Intelligent Nano-Fertilizer Management System which integrates IoT, cloud computing, and machine learning to enable precise and sustainable nutrient management. IoT-based soil and crop sensors continuously monitor key parameters such as moisture, pH, temperature, and nutrient concentration. The collected data are transmitted via wireless communication modules (Wi-Fi, LoRa, or GSM) to a cloud-based processing unit, where the information is stored and preprocessed. A machine learning module (using Random Forest, SVR, and ANN algorithms) analyzes the data to predict realtime nutrient requirements. Based on these predictions, the decision support system generates optimized fertilizer recommendations, which are then communicated to an automated dispensing unit or directly to the farmer through a mobile interface. b. Data Acquisition Field data were collected from test plots cultivated with maize and paddy over two cropping seasons. Each plot was equipped with IoT sensors capturing soil temperature, moisture, pH, electrical conductivity, and light intensity at hourly intervals. Nano-fertilizers (ZnO, Fe₂O₃, and NPK nano formulations) were applied in controlled doses based on model recommendations (Masenya 2024). c. Machine Learning Model Development Three supervised ML algorithms were implemented: 1) Random Forest (RF) for feature importance and nonlinear prediction. 2) Support Vector Regression (SVR) for continuous nutrient requirement estimation. 3) Artificial Neural Network (ANN) for multi-parameter optimization. The models were trained using 70% of the dataset and validated on the remaining 30%. Evaluation metrics included Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Coefficient of Determination (R²) (Elashmawy et al. 2023). IV. RESULTS AND DISCUSSION a. Sensor Data Trends Continuous IoT monitoring provided detailed temporal variations in soil and environmental parameters. Strong correlations were observed between soil moisture, temperature, and nutrient uptake rate, validating the need for dynamic fertilizer adjustment. Journal of Dalian University of Technology | ISSN: 1000-8608 Volume 32, Issue 11, 2025| https://jdut.net/ | Page No-310 TABLE I SENSORS AND IOT MODULES INTEGRATED IN THE INFMS Parameter Sensor Type Model/Specification Range Interface Purpose Soil Moisture Capacitive Sensor SEN0193 0–100% Analog Soil water content Soil pH pH Probe SEN0161 0–14 Analog Acidity measurement Soil Temperature Digital Thermometer DS18B20 -55°C to 125°C 1-Wire Temperature monitoring TABLE II EXPERIMENTAL DESIGN AND PARAMETER DETAILS Treatment Fertilizer Type Application Rate (kg/ha) Replications T1 Conventional 100 3 T2 Nano Zn 60 3 T3 Nano NPK 50 3 T4 Nano Composite 45 3 Fig. 2 presents a comparative bar graph showing the difference in nutrient-use efficiency between conventional fertilizer management and the proposed nano-fertilizer approach. The results clearly indicate that nano-fertilizer application significantly improves nutrient uptake efficiency (about 88%) compared to conventional methods (around 65%). This improvement highlights the effectiveness of controlled nutrient release and better plant absorption enabled by nano-formulated fertilizers, leading to enhanced resource utilization and reduced nutrient losses. Fig. 2 Comparative Nutrient use Efficiency Journal of Dalian University of Technology | ISSN: 1000-8608 Volume 32, Issue 11, 2025| https://jdut.net/ | Page No-311 b. IoT System Performances The IoT sensing network continuously monitored soil and environmental parameters at 5minute intervals, transmitting data with over 97% reliability and minimal latency (<2 s) via the cloud interface. The calibration of pH, moisture, and EC sensors exhibited an accuracy of ±2%, ensuring reliable input for the ML algorithms. The automated dashboard successfully visualized real-time trends, enabling timely decisions on nutrient dosing. Compared to manual monitoring, the system reduced labor and operational time by nearly 40%, confirming its suitability for precision agriculture applications. Fig. 3 Real-Time IoT Sensor Data Trends Continuous monitoring of soil moisture, pH, and EC provided stable, high-frequency data for the ML model. The variations recorded during the cropping cycle are shown in Fig. 3, highlighting the real-time responsiveness and accuracy of the IoT sensors in capturing soil environment dynamics. c. Machine Learning Model Performance Among the tested models, the ANN achieved the highest prediction accuracy (R² = 0.96), followed by RF (R² = 0.94) and SVR (R² = 0.91). The ANN model effectively captured the nonlinear dependencies between environmental conditions and nutrient demand. TABLE III MACHINE LEARNING MODEL PARAMETERS AND EVALUATION METRICS Model Training Samples Testing Samples RMSE MAE R² Random Forest (RF) 350 150 0.035 0.028 0.94 Support Vector Regression (SVR) 350 150 0.041 0.032 0.91 Artificial Neural Network (ANN) 350 150 0.021 0.016 0.96 Journal of Dalian University of Technology | ISSN: 1000-8608 Volume 32, Issue 11, 2025| https://jdut.net/ | Page No-312 The histograms of simulated RMSE and R² values for all three machine learning models (RF, SVR, ANN) across different combinations of covariates is shown in fig. 4. The left plot shows RMSE distributions in which lower RMSE values for ANN indicate better predictive accuracy. The right plot shows R² distributions in which ANN again exhibits consistently higher R² values, suggesting stronger goodness of fit compared to RF and SVR. Fig. 4 Histograms of RMSE and R2 values for all machine learning models Fig. 5 Density scatter plots showing the relationship between predicted and observed yield values for Random Forest (RF), Support Vector Regression (SVR), and Artificial Neural Network (ANN) models using different covariate combinations. The diagonal dashed line represents the ideal 1:1 agreement between observed and predicted values. The density patterns indicate that most data points cluster near the 1:1 line, showing strong model accuracy and reliable prediction performance across varying covariate combinations. Fig. 5 Density Scatter Plots vs. Observed Yield for ML models d. Impact on Crop Yield and Efficiency The ML-guided nano-fertilizer application resulted in: • 28–32% improvement in nutrient use efficiency. • 15–20% increase in crop yield. • 22–25% reduction in nitrate leaching and fertilizer wastage. Journal of Dalian University of Technology | ISSN: 1000-8608 Volume 32, Issue 11, 2025| https://jdut.net/ | Page No-313 These findings demonstrate the environmental and economic benefits of integrating nanofertilizers with intelligent decision systems. Fig. 6 show the relationship between actual and predicted nutrient requirements using three machine learning models—Random Forest (RF), Support Vector Regression (SVR), and Artificial Neural Network (ANN). All models closely follow the ideal 1:1 line, indicating strong predictive accuracy. Among them, the RF model demonstrates slightly higher consistency and better alignment with the ideal fit, suggesting its superior capability in capturing complex nonlinear nutrient interactions. Overall, the figure confirms the robustness and reliability of the ML models for real-time nutrient prediction in precision agriculture. Fig. 6 Predicted vs. actual nutrient requirement using different machine learning models Fig. 7 presents the correlation matrix of key soil and environmental parameters derived from IoT sensors. The heatmap illustrates the degree of association among soil pH, moisture, temperature, electrical conductivity (EC), and nutrient uptake rate. Stronger correlations (positive or negative) are represented by darker color intensities, highlighting interdependencies that influence nutrient dynamics and fertilizer efficiency. The results indicate that soil moisture and temperature exhibit a strong positive relationship with nutrient uptake, whereas pH and EC show inverse trends under certain conditions. This analysis provides insights into how multi-sensor data can improve real-time nutrient management decisions using the proposed IoT–ML system. Journal of Dalian University of Technology | ISSN: 1000-8608 Volume 32, Issue 11, 2025| https://jdut.net/ | Page No-314 Fig. 7 Correlation Matrix of Soil and Environmental Variables Fig. 8 illustrates the comparative performance of the conventional fertilizer management system and the proposed Intelligent Nano-Fertilizer Management System (INFMS), which integrates nano-formulated fertilizers with IoT and machine learning technologies. The results reveal that the INFMS achieved a substantial improvement in nutrient use efficiency (90%) compared to conventional practices (70%), indicating more precise nutrient uptake and minimal losses. Similarly, crop yield increased from 4.5 t/ha under conventional management to 5.3 t/ha with INFMS, demonstrating enhanced plant growth and productivity. Moreover, nitrate leaching was significantly reduced from 45 mg/L to 25 mg/L, highlighting the environmental advantage of the proposed system in minimizing nutrient runoff and groundwater pollution. Fig. 8 Nutrient Use Efficiency and Yield Comparison Journal of Dalian University of Technology | ISSN: 1000-8608 Volume 32, Issue 11, 2025| https://jdut.net/ | Page No-315