Authors: Ambuj Kumar Misra
Abstract: Contemporary agriculture operates under the compounding pressures of climate variability, resource scarcity, and an expanding global population. This paper introduces a comprehensive Data-Driven Agricultural Decision Support System (DADSS) that harnesses decades of historical crop production records, real-time sensor telemetry, and satellite-derived remote sensing imagery to generate actionable, site-specific recommendations for farm management. Four machine learning architectures — Random Forest (RF), Long Short-Term Memory (LSTM) networks, Gradient Boosting Machines (GBM), and Support Vector Machines (SVM) — were trained and benchmarked against a baseline linear regression model across six major U.S. cropping regions spanning the years 2000 to 2022. The hybrid DADSS framework, which fuses LSTM temporal modeling with gradient-boosted ensemble predictions, achieved an overall forecasting accuracy of 92.5%, outperforming all individual baselines. Results confirm that data-driven advisory tools can meaningfully reduce input costs, improve yield stability, and bolster farmers' adaptive capacity in the face of climate uncertainty. The system architecture, feature engineering pipeline, validation results, and policy implications are discussed in detail.
DOI: http://doi.org/10.5281/zenodo.20426302
