Soil Nutrient Prediction for Precision Agriculture with ML Algorithms

Abstract

Soil nutrient prediction is critical for precision agriculture, enabling better crop management and sustainable farming practices. Accurate predictions can optimize fertilizer use, reduce environmental impact, and enhance yield. Traditional models often struggle with the complex relationships in soil data, requiring advanced machine learning techniques. This study aims to compare the performance of two advanced boosting algorithms, CatBoost and LightGBM, for soil nutrient prediction using the Crop Recommender Dataset with Soil Nutrients from Kaggle. The dataset was preprocessed, including normalization, handling missing values, and feature encoding where necessary. CatBoost and LightGBM were trained and optimized for regression tasks to predict soil nutrient levels. Performance was evaluated using metrics such as RMSE, MAE, and R². CatBoost achieved an RMSE of 1.85 and an R² of 94.67%, while LightGBM recorded an RMSE of 1.92 and an R² of 94.11%. Both models demonstrated high accuracy, with CatBoost slightly outperforming LightGBM. The study highlights the effectiveness of boosting algorithms in soil nutrient prediction, with CatBoost proving slightly superior in terms of accuracy and interpretability. These findings emphasize the potential of ML in advancing precision agriculture.