Transparent deep learning and data-driven visual intelligence framework for robust crop recommendation in precision agriculture

Agriculture faces increasing challenges due to climate variability, soil degradation, and unpredictable weather conditions, necessitating intelligent and adaptive crop recommendation systems for sustainable farming. Traditional methods, such as Decision Trees, Random Forest, Support Vector Machines (SVM), and XGBoost, rely on static soil attributes, failing to account for long-term climate variations and lacking explainability, limiting their adoption in precision agriculture. Similarly, deep learning models like Bi-LSTM, GRU, and XAI-CROP have been explored, but they struggle with capturing complex soil-climate dependencies and ensuring transparency in decision-making. To address these challenges, this study proposes a Transformer-GRU-FNN hybrid model, a pioneering Transparent Machine Learning and Data-Driven Visual Intelligence Framework that integrates climate-aware deep learning with SHAP-based interpretability. The Transformer encoder captures intricate soil-climate interactions, GRU models long-term climate fluctuations, and FNN predicts optimal crops based on fused feature representations. Implemented using TensorFlow and PyTorch, the model is trained on the Crop Recommendation Dataset and Indian Agriculture and Climate Dataset, ensuring real-time adaptability and explainable decision-making. Comparative analysis shows that the proposed model achieves R 2 = 0.99, RMSE = 0.0321, and MAE = 0.0201, significantly outperforming Bi-LSTM (R 2 = 0.98, RMSE = 0.0424), GRU (R 2 = 0.94, RMSE = 0.0369), and XAI-CROP (R 2 = 0.94, RMSE = 0.9601). With a 6% improvement in accuracy over existing models, this approach bridges the gap between AI transparency and agricultural sustainability, providing climate-resilient, explainable, and high-performance crop recommendations to revolutionize precision agriculture.