An Augmented Reality Enabled Mobile Decision Support System for Intelligent Crop Recommendation and Yield Estimation

• Validated NPK-climate preprocessing for robust agricultural ML inference. • Neural-network engine for joint crop suitability and yield prediction. • Edge-based real-time inference via ONNX mobile deployment. • AR-driven predictive crop visualization in physical farmland context. • Unified predictive-immersive mobile agricultural intelligence framework. Crop selection is a critical determinant of agricultural productivity, resource efficiency, and long-term sustainability, as it directly influences yield potential and input optimization. However, traditional crop planning approaches- primarily based on regional practices or experiential knowledge often fail to account for field-specific variability and dynamically changing environmental conditions. Although existing digital crop recommendation systems employ predictive analytics, their outputs are typically presented in textual or numerical formats; consequently, their interpretability and practical usability in real farming contexts remain limited. To overcome these limitations, this study proposes an interactive Augmented Reality (AR) enabled mobile decision-support framework that integrates predictive intelligence with immersive visualization. Specifically, the system combines soil macronutrient parameters Nitrogen (N), Phosphorus (P), and Potassium (K) with geolocation-derived environmental variables, including temperature and rainfall, to evaluate crop suitability using a neural-network-based model. Furthermore, the trained model is deployed locally via an ONNX architecture; therefore, real-time inference can be achieved without reliance on cloud infrastructure. In addition, a linear yield estimation module is incorporated to support planning decisions based on land area and crop-specific productivity coefficients. Moreover, the AR interface enables in situ visualization of predicted crops and yield insights within real farmland environments, thereby enhancing contextual understanding. The model is evaluated on a structured dataset comprising rice, maize, and cotton, demonstrating effective performance under resource-constrained conditions. Nevertheless, limitations related to dataset scope and simplified yield modeling remain, indicating directions for future research.