Scholarlink

<p>Agricultural runoff and seasonal variations significantly influence river water quality, yet existing monitoring systems often lack the accuracy, adaptability, and cost-efficiency required for continuous assessment. This study introduces a novel hybrid framework, Multi-Linear Regression and Multi-class XGBoost-based Water Quality Monitoring (MRMX-WQM), integrating statistical and machine learning approaches for improved prediction reliability. Real-time data on pH, temperature, nitrate, salinity, ammonia, and turbidity were collected from five monitoring sites using a low-cost Raspberry Pi-based sensor system. Multi-Linear Regression captured fundamental linear dependencies among parameters, while Multi-class XGBoost modeled complex nonlinear relationships, enhancing predictive performance. The proposed system includes an automated alert mechanism to notify stakeholders when pollutant thresholds are exceeded. Experimental evaluation across summer and winter seasons in agricultural catchments demonstrated that MRMX-WQM outperformed conventional machine learning methods, achieving up to 6.43% improvement in R² and reductions of 2.24% in RMSE. MRMX-WQM consistently achieved better results than the other approaches, with performance improvement of 3.21% over SVM, 4.32% over KNN, and an impressive 7.89% advantage over Naïve Bayes. This cost-effective, scalable, and seasonally adaptive solution can be replicated for diverse water bodies, offering a practical tool for sustainable water quality management.</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

Machine learning for river water quality monitoring: assessing seasonal and agricultural influences

  • 期刊论文

  • 发表日期：

    2025年10月25日

查看全文
  • Ganesh Babu R,
  • Geetha T S,
  • Ramachandra Reddy K,
  • Kavin Kumar K

摘要

Agricultural runoff and seasonal variations significantly influence river water quality, yet existing monitoring systems often lack the accuracy, adaptability, and cost-efficiency required for continuous assessment. This study introduces a novel hybrid framework, Multi-Linear Regression and Multi-class XGBoost-based Water Quality Monitoring (MRMX-WQM), integrating statistical and machine learning approaches for improved prediction reliability. Real-time data on pH, temperature, nitrate, salinity, ammonia, and turbidity were collected from five monitoring sites using a low-cost Raspberry Pi-based sensor system. Multi-Linear Regression captured fundamental linear dependencies among parameters, while Multi-class XGBoost modeled complex nonlinear relationships, enhancing predictive performance. The proposed system includes an automated alert mechanism to notify stakeholders when pollutant thresholds are exceeded. Experimental evaluation across summer and winter seasons in agricultural catchments demonstrated that MRMX-WQM outperformed conventional machine learning methods, achieving up to 6.43% improvement in R² and reductions of 2.24% in RMSE. MRMX-WQM consistently achieved better results than the other approaches, with performance improvement of 3.21% over SVM, 4.32% over KNN, and an impressive 7.89% advantage over Naïve Bayes. This cost-effective, scalable, and seasonally adaptive solution can be replicated for diverse water bodies, offering a practical tool for sustainable water quality management.