AgroSage: A GA-Tuned Random Forest Framework for Smart Disease Diagnosis and Fertilizer Recommendation in Vegetable Crops

Precision agriculture requires scalable, interpretable, and deployable solutions ensuring consistent delivery of crop health assessments and nutrient management guidance. In this work, a two-stage plant disease classification and fertilizer recommendation system for tomato, potato, and pepper plants is proposed and named AgroSage, utilizing a GA-tuned Random Forest (RF) model. Unlike deep learning methods, which are computationally expensive and demand a large amount of data, AgroSage makes use of hand-engineered features such as color histograms, Local Binary Patterns (LBP), and shape descriptors derived from leaf images to build an efficient and interpretable classifier. The system is designed to detect five common diseases: early blight, late blight, bacterial wilt, anthracnose, and leaf curl virus. After a disease is detected, a second GA-optimized random forest-based model takes the crop type and soil macronutrient (N, P, K) levels as input to recommend the type and quantity of fertilizer that should be applied and the application method. Both models are incorporated into a lightweight web interface that allows for real-time inference and multilingual input, as well as offline caching. Stratified cross-validation verifies classification accuracy over 95% and fertilizer application accuracy at 95.2%. Customized forms of AgroSage can provide site-specific, disease-specific, and soil-specific recommendations to help farmers achieve healthy and sustainable crop yields, reduce chemical overuse, and bolster the resilience of precision agriculture systems in resource-constrained farming areas.