Crop Yield Prediction Using Machine Learning: A Random Forest-Based Crop Recommendation System

Abstract

Agricultural productivity depends on the interaction of soil nutrients, weather, water availability, and crop-specific growing requirements. Selecting an unsuitable crop can reduce production, waste fertilizers and irrigation, and increase financial risk. This paper presents a machinelearning crop recommendation framework developed from the accompanying crop-yield project. The system accepts nitrogen, phosphorus, potassium, temperature, humidity, soil pH, rainfall, and water-level information, applies data cleaning and feature preparation, and compares Logistic Regression, Decision Tree, and Random Forest classifiers. The implementation uses a 70:30 train-test partition and a Random Forest with 500 entropy-based trees. The reported experimental comparison gives accuracies of 93.63%, 95.45%, and 97.87% for Logistic Regression, Decision Tree, and Random Forest, respectively. Random Forest is serialized and integrated with a Flask interface that receives agricultural parameters and returns the recommended crop. The architecture connects dataset preparation, model selection, prediction, and user-facing deployment in one workflow. Future improvements include live weather and IoT sensor feeds, regional calibration, remote-sensing variables, multilingual interaction, and crop-specific yield regression.