AI-POWERED CROP YIELD PREDICTION AND OPTIMIZATION

Abstract

Agriculture faces increasing challenges due to climate variability, plant diseases, soil degradation, and inefficient resource utilization, necessitating intelligent data-driven solutions for sustainable food production. This work presents an AI-powered agricultural prediction framework that integrates plant disease detection, crop recommendation, yield prediction, rainfall/climate analysis, and soil moisture estimation using machine learning, deep learning, ensemble learning, and explainable AI techniques. For plant disease prediction, deep convolutional neural networks including ResNet50, VGG16, Xception, NASNet, and a hybrid ensemble model are employed with extensive image augmentation. Crop recommendation is addressed using Random Forest, Decision Tree, KNN, XGBoost, AdaBoost, and a Voting Classifier. Yield prediction, rainfall/climate analysis, and soil moisture estimation are modeled using both classification and regression approaches with the same set of ensemble-based algorithms. Explainable AI techniques such as SHAP, LIME, and GradCAM are utilized to enhance transparency and interpretability of predictions. Experimental evaluation demonstrates that the Voting Classifier achieves superior performance for crop recommendation with an accuracy of 98.6%. Yield prediction shows the best performance using a voting-based ensemble regressor with an R² score of 91.3%. Soil moisture prediction attains high accuracy using a Voting ensemble model with 97.7% performance. Plant disease detection achieves 99.8% accuracy using a hybrid deep learning ensemble, while rainfall analysis is effectively handled using a Voting Classifier. The integrated framework supports precise decision-making for crop management, irrigation scheduling, and disease control.