Scholarlink

<p>Accurate and timely assessment of river water quality is critical for sustainable water management. Traditional water quality index (WQI) calculation methods, however, are often labor-intensive and limited in their capacity to handle complex, multivariate datasets. This study addresses these limitations by developing a machine learning (ML) framework to predict WQI in Thailand’s lower Mun river basin. Comparative evaluation of individual models−extreme gradient boosting (XGBoost), random forest (RF), support vector regression (SVR), and decision tree (DT)-along with their hybrid combinations demonstrated that the XGBoost-RF hybrid achieved superior predictive performance (training: root mean squared error (RMSE) = 0.1095, coefficient of determination (R<sup>2</sup>) = 0.9998; testing: RMSE = 2.0420, R<sup>2</sup> = 0.9332). To quantify prediction uncertainty, the lower upper bound estimation (LUBE) method was integrated, yielding robust prediction intervals with high coverage (prediction interval coverage probability (PICP) = 96.7%) and satisfactory width (prediction interval normalized root-mean-square width (PINRW) = 0.287). Explainable artificial intelligence (XAI) techniques employing Shapley additive explanations (SHAP) values identified biochemical oxygen demand (BOD) and ammonia (NH<sub>3</sub>) as primary drivers of WQI variability, thereby providing actionable insights for targeted pollution control strategies. The developed methodology is computationally efficient, interpretable, and scalable, rendering it suitable for real-time water quality monitoring systems. This research presents an integrated approach that simultaneously enhances prediction accuracy and model transparency in WQI forecasting under practical data constraints. </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

Lower and upper bound estimation (LUBE) for predicting the surface water quality in the lower Mun river basin using machine learning algorithms

  • 期刊论文

  • 发表日期：

    2025年7月8日

查看全文
  • Chawisa Chawishborwornworng,
  • Monchai Chansiri,
  • Santamon Luanwuthi,
  • Chakkrit Umpuch,
  • Channarong Puchongkawarin

摘要

Accurate and timely assessment of river water quality is critical for sustainable water management. Traditional water quality index (WQI) calculation methods, however, are often labor-intensive and limited in their capacity to handle complex, multivariate datasets. This study addresses these limitations by developing a machine learning (ML) framework to predict WQI in Thailand’s lower Mun river basin. Comparative evaluation of individual models−extreme gradient boosting (XGBoost), random forest (RF), support vector regression (SVR), and decision tree (DT)-along with their hybrid combinations demonstrated that the XGBoost-RF hybrid achieved superior predictive performance (training: root mean squared error (RMSE) = 0.1095, coefficient of determination (R2) = 0.9998; testing: RMSE = 2.0420, R2 = 0.9332). To quantify prediction uncertainty, the lower upper bound estimation (LUBE) method was integrated, yielding robust prediction intervals with high coverage (prediction interval coverage probability (PICP) = 96.7%) and satisfactory width (prediction interval normalized root-mean-square width (PINRW) = 0.287). Explainable artificial intelligence (XAI) techniques employing Shapley additive explanations (SHAP) values identified biochemical oxygen demand (BOD) and ammonia (NH3) as primary drivers of WQI variability, thereby providing actionable insights for targeted pollution control strategies. The developed methodology is computationally efficient, interpretable, and scalable, rendering it suitable for real-time water quality monitoring systems. This research presents an integrated approach that simultaneously enhances prediction accuracy and model transparency in WQI forecasting under practical data constraints.