Scholarlink

<p>Solar disinfection (SODIS) is a low-cost and sustainable household water treatment method endorsed by the WHO. Despite its widespread use, no machine learning (ML) models have been developed to predict the <i>E. coli</i> inactivation rate under SODIS conditions. This study addresses this gap by developing prediction models for the bacterial inactivation rate (k) using ML algorithms, including ANN, MLR, SVM, and RF. The models are based on 30 experimental datasets from SODIS with H<sub>2</sub>O<sub>2</sub> trials conducted in PET bottles and plastic bags during monsoon and winter seasons in Bangladesh. Input variables include water temperature, solar irradiance, turbidity, and dissolved oxygen. Data augmentation techniques (datasets of 5, 10, 30, and 50) were applied to enhance model performance, which was evaluated using R<sup>2</sup>, MAE, RMSE, and MAPE. The RF model generated the best outcomes with an R<sup>2</sup> of 0.98, followed by ANN and SVM (R<sup>2</sup> = 0.98), and MLR (R<sup>2</sup> = 0.91). Moreover, ANN models using the Tanh activation function outperformed ReLU and Sigmoid. The study demonstrates that significant ML models can be developed with limited data through augmentation techniques, offering a reliable tool to predict <i>E. coli</i> inactivation rates. These findings enhance the application and global reliability of SODIS.</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

Comparison of machine learning models with data augmentation techniques for predicting E. coli die-off rate in solar disinfection

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年8月28日

查看全文
  • Md. Habibur Rahman Bejoy Khan,
  • Md. Rezaul Karim,
  • Nafisa Anjum Rimi,
  • Mastura Morshed Nawmi,
  • Fuad Bin Nazrul,
  • Amimul Ahsan,
  • Tahmeed Ahmed,
  • Monzur Alam Imteaz,
  • Mohammad T. Alresheedi

摘要

Solar disinfection (SODIS) is a low-cost and sustainable household water treatment method endorsed by the WHO. Despite its widespread use, no machine learning (ML) models have been developed to predict the E. coli inactivation rate under SODIS conditions. This study addresses this gap by developing prediction models for the bacterial inactivation rate (k) using ML algorithms, including ANN, MLR, SVM, and RF. The models are based on 30 experimental datasets from SODIS with H2O2 trials conducted in PET bottles and plastic bags during monsoon and winter seasons in Bangladesh. Input variables include water temperature, solar irradiance, turbidity, and dissolved oxygen. Data augmentation techniques (datasets of 5, 10, 30, and 50) were applied to enhance model performance, which was evaluated using R2, MAE, RMSE, and MAPE. The RF model generated the best outcomes with an R2 of 0.98, followed by ANN and SVM (R2 = 0.98), and MLR (R2 = 0.91). Moreover, ANN models using the Tanh activation function outperformed ReLU and Sigmoid. The study demonstrates that significant ML models can be developed with limited data through augmentation techniques, offering a reliable tool to predict E. coli inactivation rates. These findings enhance the application and global reliability of SODIS.