Studi Klasifikasi Presisi Kesesuaian Lahan Pertanian Menggunakan Algoritma Extreme Gradient Boosting (XGBoost)

Abstract

The application of artificial intelligence technology in the agricultural sector is the main foundation in the paradigm shift towards sustainable precision agriculture. This study presents a comprehensive analysis of the application of the Extreme Gradient Boosting (XGBoost) algorithm to predict the suitability of crop types based on soil chemical characteristics and macro-environmental conditions. Model evaluation was conducted using the benchmark dataset Crop Recommendation Dataset accessed through the Kaggle platform. This dataset has a perfect class balance with a total of 2,200 samples evenly divided into 22 agricultural commodities. The developed predictive model evaluates seven soil and climate biophysical parameters, namely nitrogen, phosphorus, potassium, air temperature, relative humidity, soil acidity (pH), and rainfall intensity. The test results show that the XGBoost algorithm ranks top in classification accuracy with values ​​ranging from 99.31% to 99.77%, surpassing other ensemble algorithms such as Random Forest as well as traditional models such as Decision Tree and Naive Bayes. Feature contribution analysis demonstrates that climate parameters (rainfall and humidity) act as primary ecological filters at the macro-level, while soil macronutrient (NPK) ratios serve as secondary determinants at the crop-specific micro-level. Overall, this boosting-based ensemble approach offers high accuracy and robustness to data outliers, making it a highly reliable agronomic decision-making tool for supporting sustainable land productivity.