Comparative evaluation of decision tree and random forest models for multi-class water quality classification of the Yamuna river

Abstract

Monitoring the quality of water in rivers is important for environmental sustainability. This study presents a comparison of Decision Tree (DT) and Random Forest (RF) classifiers for multi-class classification of the water quality of the Yamuna River on the basis of temperature, pH, electrical conductivity, BOD, and fecal coliform concentration. To enhance the robustness of the findings, noise of ± 5% Gaussian was injected to simulate real-world variability. Moreover, the impact of class imbalance was addressed using SMOTE oversampling. Model performance was test by stratified fivefold cross validation. The accuracy of decision tree model was found to be 76.30%, precision 76.28%, recall 76.31 and f1 score 72.55% The accuracy, precision, recall and F1-score of Random Forest classifier was 78.00%, 74.59%, 78.00% and 74.41%. The results show ensemble learning improves the stability of classification and the reliability of prediction for multi-class river water quality assessment. The findings of the comparative study indicate the effective use of tree-based algorithms for river monitoring.