Data-Driven Crop Recommendation Using Machine Learning: A Comparative Analysis

Abstract

This research presents a comprehensive comparative analysis of machine learning algorithms for agricultural decision support systems using the Crop Recommendation Dataset. Four supervised learning algorithms—Random Forest (RF), Decision Tree (DT), Support Vector Machine (SVM), and K-Nearest Neighbors (KNN)—were implemented, trained, and rigorously evaluated under consistent conditions. The manually optimized Random Forest model achieved the highest performance, attaining an overall accuracy of 99.32%, outperforming Decision Tree (97.95%), SVM (97.73%), and KNN (95.68%). Each model was tuned through controlled hyperparameter optimization to ensure fair comparison. The Random Forest model revealed critical feature importance insights, highlighting rainfall (22.7%), humidity (21.1%), and potassium levels (18.1%) as dominant agronomic parameters influencing prediction outcomes. This study establishes a performance hierarchy among the tested algorithms, formulates algorithm selection guidelines, and provides practical recommendations for the development of intelligent crop recommendation systems. The findings contribute to the growing field of agricultural informatics and precision agriculture by demonstrating the feasibility of integrating ML-driven decision models into real-world farming environments.