Volume 1, Issue 2 — Year 2024 — Article e100018

Ramin Vafaei Poursorkhabi¹; Alireza Naseri²

1. Department of Civil Engineering, Tabriz Branch, Islamic Azad University, Tabriz 5158913791, Iran
2. Robotics & Soft Technologies Research Center, Tabriz Branch, Islamic Azad University, Tabriz 5158913791, Iran

This study develops Adaptive Neuro-Fuzzy Inference Systems (ANFIS) and Gaussian Process Regression (GPR) models to estimate sediment transport in circular channels used in water and wastewater systems and compares their performance with two widely applied machine-learning techniques: Artificial Neural Networks (ANN) and Support Vector Machines (SVM). The modeling framework incorporates key hydraulic and sediment parameters, including shear velocity, hydraulic radius, particle-size ratios, and mobility indices, to explore their nonlinear influence on sediment discharge. The results show that ANFIS and GPR consistently outperform ANN and SVM, offering more stable and reliable predictions across both smooth and rough bed conditions. ANFIS demonstrates the highest accuracy due to its hybrid learning structure, while GPR provides strong generalization with minimal overfitting. ANN delivers moderate accuracy but is highly sensitive to data variability, whereas SVM tends to underpredict sediment transport, particularly at higher flow intensities. The best-performing models for ANFIS and GPR were achieved using the input set (u*, d₅₀/R, λs, and Fd), highlighting the strong interaction between flow intensity and particle mobility. Sensitivity analysis identifies the particle Froude number (Fd) as the most influential parameter. Additionally, smoother channel beds improve prediction accuracy for all models, while increased roughness reduces sediment transport efficiency and introduces higher predictive uncertainty.

Sediment transport, Closed-conduit flow, Data-driven modeling, Flow–particle interactions, Bed surface condition