Performance Evaluation of Machine Learning Models for Soil Fertility Classification Based on the Indian Soil Fertility Dataset

Abstract

Rice farming productivity worldwide has been declining due to improper soil management practices, including excessive chemical fertilizer use and irregular irrigation. The main challenge lies in accurately classifying soil fertility levels to support optimal land use and reduce resource waste, especially when dealing with imbalanced datasets. This study aims to compare the performance of single classifiers and ensemble classifiers in classifying soil fertility. The single classifiers used include K-Nearest Neighbor (KNN), Naive Bayes, Decision Tree, Support Vector Machine (SVM), and Artificial Neural Network (ANN), while the ensemble classifiers include Random Forest and XGBoost. The Indian Soil Fertility Dataset, obtained from Kaggle, contains 880 samples with 12 features and 1 output class. The research methodology involved data acquisition, preprocessing, data splitting, standardization, and classification, with performance evaluation conducted using a confusion matrix. The results show that ensemble classifiers, particularly Random Forest and XGBoost, outperform single classifiers in imbalanced datasets, achieving accuracy, precision, recall, and F1-score values exceeding 92%-95% across all split scenarios. The findings conclude that Random Forest and XGBoost can serve as reliable models for assisting farmers and agricultural experts in evaluating soil conditions, minimizing unnecessary fertilizer usage, and improving rice farming productivity globally.

Keywords

• Classification
• Ensemble Classifier
• Machine Learning
• Single Classifier
• Soil Fertility Classification