AgroCare: An Integrated AI-Driven Agricultural Intelligence Platform for Crop Recommendation, Soil Health Diagnosis, and Smart Farming Optimization

Authors:

Sakthi A, Sathish K, Sujith Kumar K, Ganesh Kumar P, Manikandan R 

Abstract—Agriculture remains the backbone of economies in developing nations, yet farmers continue to face challenges arising from inadequate decision-support tools, unpredictable soil conditions, improper crop rotation, and volatile market prices. This paper presents AgroCare, an end-to-end artificial intelligence-driven agricultural intelligence platform that integrates five core modules: crop recommendation, soil health classification, crop rotation planning, market price prediction, and plant disease detection. AgroCare employs stacking ensemble methods combining Random Forest (RF), XGBoost (XGB), and LightGBM (LGBM) for structured tabular tasks, and EfficientNetV2S transfer learning for image-based plant disease detection. Experimental evaluations demonstrate outstanding performance across modules: the crop recommendation model achieves 99.77% accuracy across 22 crop classes, the soil health classifier attains 100% accuracy over five health categories, and the crop rotation planner achieves 94.50% accuracy. Market price prediction yields an R² score of 0.8399 using XGBoost regression, while plant disease detection reaches 36.53% accuracy at Epoch 1 of ongoing fine-tuning. All classification models attain a macro-averaged AUC-ROC of 1.00. The system is deployed as a Flask-based web application, offering an accessible, browser-driven interface for farmers and agronomists. AgroCare demonstrates that multi-task ensemble intelligence can substantially elevate agricultural decision-making and support sustainable farming practices.

Index Terms—Precision agriculture, ensemble learning, crop recommendation, soil health, plant disease detection, XGBoost, LightGBM, EfficientNetV2S, stacking ensemble, agricultural AI