Implementation of Hybrid model for Enhanced Crop Yield Prediction using Explainable AI for Precision Agriculture

Crop yield prediction is important for global food security, sustainable farming, and efficient resource management in agriculture. Inaccuracy and inherent opacity are the main issue for accurate prediction. To solve the problems of deep learning models a hybrid model has been developed for better prediction. This research works to improve agricultural decision-making through advanced predictive modelling. In this research work SHAP (SHapley Additive exPlanations) and Grad-CAM (Gradient-weighted Class Activation Mapping), have been employed. SHAP analysis helps us to understand the input feature impact of soil moisture at middle level growth and helps us to analyse the over fertilization that is no longer increasing the crop yield rather reducing it. This work shows integration of deep learning and explainable AI for data-driven decisions on irrigation process, usage of fertilizers, and strategies to improve the decision making of farmers and thus heading towards precision farming. A hybrid model that considers dynamic and static characteristics to forecast crop yields. The model employs Conv1D layers for finding short-term patterns and Bidirectional LSTM layers for long-term dependencies, along with dedicated dense layers for static features. The overall $\mathbf{R}^{\boldsymbol{\wedge}} \mathbf{2}$ is $\mathbf{0. 8 1}$ and $\mathbf{R}^{\boldsymbol{\wedge}} \mathbf{2}$ for specific crop types (wheat, soybean, corn, and rice) is always 0.91 and above, therefore outperforming models based on exclusive use of dynamic or static characteristics by a wide margin.