Scholarlink

<p>Access to healthy and potable water is a fundamental human requirement, as well as a foundation for public health and environmental sustainability. To address this need, the present study employs a water quality classification approach to distinguish between potable and non-potable water based on a range of physicochemical parameters. Several machine-learning models, including ExtraTrees, LightGBM, AdaBoost, KNN, XGBoost, Decision Tree, and Bagging, were employed as candidate techniques for predicting water drinkability. To enhance the stability of the classification models, the SMOTE was applied to address class imbalance between potable and non-potable samples, while the GJO approach was employed to optimize each model’s hyperparameters. From the results, the ExtraTrees classifier showed moderate predictive factors, 0.7150 accuracy and 0.7179 precision, in which the integration of the SMOTE, along with GJO hyperparameter optimization, helped in improving generalization stability and robustness. Furthermore, the importance of water quality features was assessed using SHAP and Permutation analysis based on ExtraTrees classifier. This analysis revealed that pH and sulfate were more relevant than other listed features, which implies that these parameters function as high‑leverage control variables within the system. Therefore, these results encourage the evolution of automatic water quality surveillance, which opens the prospects of practical application to the area of public health.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Optimizing potable water prediction by advanced analytical models with feature insights

  • 期刊论文

  • 发表日期：

    2026年5月26日

查看全文
  • Jianchao Zhu,
  • Yin Wang,
  • Rui Gao

摘要

Access to healthy and potable water is a fundamental human requirement, as well as a foundation for public health and environmental sustainability. To address this need, the present study employs a water quality classification approach to distinguish between potable and non-potable water based on a range of physicochemical parameters. Several machine-learning models, including ExtraTrees, LightGBM, AdaBoost, KNN, XGBoost, Decision Tree, and Bagging, were employed as candidate techniques for predicting water drinkability. To enhance the stability of the classification models, the SMOTE was applied to address class imbalance between potable and non-potable samples, while the GJO approach was employed to optimize each model’s hyperparameters. From the results, the ExtraTrees classifier showed moderate predictive factors, 0.7150 accuracy and 0.7179 precision, in which the integration of the SMOTE, along with GJO hyperparameter optimization, helped in improving generalization stability and robustness. Furthermore, the importance of water quality features was assessed using SHAP and Permutation analysis based on ExtraTrees classifier. This analysis revealed that pH and sulfate were more relevant than other listed features, which implies that these parameters function as high‑leverage control variables within the system. Therefore, these results encourage the evolution of automatic water quality surveillance, which opens the prospects of practical application to the area of public health.