From Sensors to Insights: Leveraging Intelligent Data Mining for Precision Farming and Decision Support

Climate change, resource scarcity, and growing weather variability make sustainable farm management increasingly difficult. This study presents a precision-agriculture framework that fuses real-time Internet of Things (IoT) measurements with satellite-based environmental information to support data-driven decisions. Low-cost wireless sensor nodes record soil moisture, pH, air temperature, relative humidity, water turbidity, and total dissolved solids, and we combine these field streams with NASA Earthdata, Landsat, and Agriculture and Water Management Data Pathfinder products. We clean, validate, normalize, and align the heterogeneous data in time and space before modeling. We use four task-specific models: YOLOv5 for plant disease and pest detection, a long short-term memory (LSTM) network for short-term ultraviolet (UV) index forecasting, a CNN2D-BGRU network for weather forecasting, and an RF-LSTM-Attention model for seasonal rainfall prediction to support flood and drought assessment. Outputs are delivered through a web-based decision-support platform with live visualizations and threshold-based alerts. The detector achieved 92.1% accuracy (95% CI 87.4–95.2%), 94.7% recall, and a Matthews correlation coefficient of 0.84. The UV model achieved a mean absolute error of 0.249 index units, and the rainfall model a mean absolute percentage error of 9.49%. A prevalence-aware analysis shows that at 10% field disease prevalence, only half of image-based alerts would be true positives, and that using IoT microclimate conditions as a prior can raise alert precision to about 79%. These results support multi-source fusion for precision agriculture, while sensor reliability, missing data, extreme weather, and model generalization remain open challenges.