Scholarlink

<p>In the context of steep mountainous channels, the measurement of friction factor (<i>f</i>) is one of the crucial parameters for understanding hydraulics and predicting head loss due to hydraulic resistance. Also, laboratory experiments pose limitations on the predictability of “<i>f</i>” because of the traditional forecasting methods. To address these challenges, advanced machine learning (ML) techniques offer promising solutions. In this study, eight ML models, such as Random Forest, Bagging, Linear Regression, Ridge, Lasso, K-Nearest Neighbors, AdaBoost, and Gradient Boosting, are employed to predict “<i>f</i>” using distinct input features. By following that, performance evaluation and various error metrics are computed to determine the accuracy of the ML models. Graphical analyses, including heatmap visualization, Taylor diagrams, sensitivity analysis, shapely additive explanations, regression error characteristic curves, and parametric analysis, provide valuable insights. Results indicate that ML models such as Random Forest and Bagging achieved high <i>R</i><sup>2</sup> values of 0.959 and 0.950, respectively, signifying strong predictive capability. These models outperform other ML approaches, demonstrating superior accuracy in predicting with minimal errors. The enhanced precision of the models emphasise significance within the realm of hydraulic engineering, particularly in the assessment of “<i>f</i>” as the experimental methods demand considerable time and resources.</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

A Novel Approach for Predicting Friction Factor in Steep Mountain Channels: An Investigation into Machine Learning Methodologies for Complex Phenomena

  • 期刊论文

  • 发表日期：

    2026年4月13日

查看全文
  • Ajaz Ahmad Mir,
  • Mahesh Patel

摘要

In the context of steep mountainous channels, the measurement of friction factor (f) is one of the crucial parameters for understanding hydraulics and predicting head loss due to hydraulic resistance. Also, laboratory experiments pose limitations on the predictability of “f” because of the traditional forecasting methods. To address these challenges, advanced machine learning (ML) techniques offer promising solutions. In this study, eight ML models, such as Random Forest, Bagging, Linear Regression, Ridge, Lasso, K-Nearest Neighbors, AdaBoost, and Gradient Boosting, are employed to predict “f” using distinct input features. By following that, performance evaluation and various error metrics are computed to determine the accuracy of the ML models. Graphical analyses, including heatmap visualization, Taylor diagrams, sensitivity analysis, shapely additive explanations, regression error characteristic curves, and parametric analysis, provide valuable insights. Results indicate that ML models such as Random Forest and Bagging achieved high R2 values of 0.959 and 0.950, respectively, signifying strong predictive capability. These models outperform other ML approaches, demonstrating superior accuracy in predicting with minimal errors. The enhanced precision of the models emphasise significance within the realm of hydraulic engineering, particularly in the assessment of “f” as the experimental methods demand considerable time and resources.