Comparative Analysis of Random Forest, Support Vector Machine and K-Nearest Neighbors for Two-Stage Crop Recommendation and Agricultural Outcome Prediction

This paper’s key contribution is a two-stage machine learning pipeline that first predicts a suitable crop from soil and environmental parameters, then predicts five agricultural outcomes (planting time, harvest time, fertilizer requirement, pest infestation level, and crop yield) for a selected crop variety, extending beyond simple crop recommendation found in most prior work. It further provides a controlled, head-to-head comparison of three algorithms (Random Forest, SVM, and KNN) on the same dataset and pipeline, which is rare in existing literature. Most importantly, the study critically examines Random Forest’s near-perfect Stage-2 scores (accuracy and R² close to 1.0) and identifies them as likely artifacts of the dataset’s deterministic structure rather than genuine model performance, a scrutiny largely absent in comparable studies that report high accuracy without question. Based on this credibility analysis, the paper concludes that KNN offers the most reliable and balanced results, while SVM performs well only for classification tasks and Random Forest requires further validation before real-world deployment—offering practical guidance for algorithm selection in precision agriculture systems.