Decision Support Systems (DSS) in precision agriculture have evolved from static rule-based engines into data-driven frameworks powered by Internet of Things (IoT) telemetry and supervised machine learning. While recent literature increasingly focuses on deep learning and volatile commodity forecasting, empirical farm-level deployments rely heavily on structured tabular telemetry—comprising soil macronutrients (N, P, K), moisture dynamics, potential of hydrogen (pH), and local microclimate variations.

This systematic review synthesizes research published between 2018 and 2026, evaluating non-deep machine learning paradigms tailored for tabular agro-telemetry across 58 peer-reviewed studies. We analyze multi-tier system architectures spanning low-power field acquisition, edge preprocessing, and predictive inference engines.

Comparative benchmarking indicates that tree-based ensembles (Random Forest, XGBoost, LightGBM) consistently outperform kernel methods and linear baselines, yielding classification accuracies between 94.2% and 99.4% for crop suitability and fertilizer recommendation while maintaining sub-millisecond inference latencies suitable for resource-constrained edge gateways. Support Vector Machines (SVM) with radial basis function (RBF) kernels offer competitive stability on small-sample regime topologies.

Finally, we highlight critical deployment challenges, including sensor calibration drift, missing data imputation under intermittent connectivity, and edge compute trade-offs, outlining viable architectural directions for resilient agro-decision frameworks

Keywords— Precision agriculture; Decision support systems; Agro-telemetry; Random Forest; XGBoost; Support vector machines; Edge computing.