Water Quality Index Prediction Using Feature Selection and Stacking Models

The growing need for real-time, accurate, and scalable water quality monitoring systems has underscored the drawbacks of traditional methods, which can be resource-intensive, inefficient, and costly. This study addresses these challenges by employing computer vision models to predict the Water Quality Index (WQI) using a dataset of diverse water quality parameters. A Chi-square test is utilized for feature extraction, followed by the application of stacking ensemble prediction techniques particular. The hybrid combination of Support Vector Machines (SVM) and XGBoost exhibited the good performance with the accuracy of 93.1% and precision, recall, and F1-score values of 0.93 on an 80:20 train/test split. Additionally, alternative stacking methods, including SVM with AdaBoost, Random Forest, and Linear Regression, were evaluated under various train/test ratios. A detailed analysis of model performance was conducted using evaluation metrics and confusion matrices to compare predicted and actual labels. The findings offer significant insights into the efficacy of the proposed methods for Water Quality Index (WQI) prediction, highlighting their potential in improvising the performance and reliability of automated water quality monitoring systems.