Interpretable Paddy Yield Prediction using Dynamic Dropout Regularized FCNN with Shap Explanations

Accurate paddy yield prediction using structured agricultural datasets is essential for planning, resource allocation, and decision support. However, many deep learning approaches for yield forecasting either rely on complex multimodal inputs or provide limited interpretability, making them difficult to trust and deploy in real-world settings. This paper presents an explainable deep learning framework based on a Feature Normalization and Dropout Regularized Fully Connected Neural Network (FN-DR-FCNN) for paddy yield prediction using tabular agronomic and environmental variables. The model integrates z-score feature normalization and a dynamic dropout mechanism to improve generalization and reduce overfitting during training. To enhance transparency, The SHapley Additive exPlanations (SHAP) are also included to measure the contribution of each input feature to the predicted yield to be able to interpret it globally and locally. The proposed approach is validated on two real-world datasets, namely the LDS Rice Dataset and the Vellore Rice Dataset. Experimental results demonstrate that FN-DRFCNN achieves strong predictive performance while providing consistent and meaningful feature attributions that support agronomic reasoning. The combination of accuracy, computational efficiency, and explainability makes the proposed framework suitable for practical decision-making in smart agriculture applications.