AI-Based Smart Agriculture: An Integrated Machine Learning Framework for Crop Recommendation, Fertilizer Recommendation and Crop Yield Prediction

Agricultural productivity in India is strongly affected by weather variability, soil degradation and the over- or under-use of fertilizers. Farmers still tend to rely on personal experience when choosing a crop, estimating its yield and deciding how much fertilizer to apply. This paper presents an integrated, web-based decision-support system that performs three tasks within a single framework: crop recommendation, fertilizer recommendation and crop yield prediction. The system takes soil nutrients (nitrogen, phosphorus and potassium), temperature, humidity, soil pH, rainfall, soil moisture, soil type and crop type as inputs. The pipeline consists of data acquisition from public repositories, cleaning and normalisation, correlation-based feature selection, and model training using Artificial Neural Networks (ANN) and Random Forest (RF). Model performance is evaluated with cross-validation using accuracy, precision, recall, F1-score and correlation-based measures. In the reported experiments, the ANN model for yield prediction reached an accuracy of 94.70% and the Random Forest model for fertilizer recommendation reached 93.94%, and the crop recommendation model [insert accuracy] was the best performer for its task. The models are packaged in a Java (JSP/Servlet, Weka, MySQL) web application that lets farmers obtain all three recommendations from one input form..