10.25676/11124/173205
Huang, Li-Jeng
Hsiao, Darn-Horng
On the regression of velocity distribution of debris flows using machine learning techniques
Mountain Scholar
2019
data analysis
debris flows
machine learning
nonlinear regression
velocity distribution
Colorado School of Mines. Arthur Lakes Library
Colorado School of Mines. Arthur Lakes Library
2019-08-21
2019-08-21
2019
eng
Text
https://hdl.handle.net/11124/173205
http://dx.doi.org/10.25676/11124/173205
born digital
proceedings (reports)
Copyright of the original work is retained by the authors.
Five machine learning techniques-- classical nonlinear regression (NLR), multi-layer perceptrons (MLP), support vector machines (SVM) with radial-basis function (RBF) kernel, k nearest neighbour (kNN) and decision tree (DT) schemes-- were applied for regression of velocity distribution along the depth of debris flows by using experimental data of steady uniform open-channel flows. Programs coded in Python and package scikit-learn were developed for machine learning analyses. Experimental results of two cases conducted and published by Matsumura and Mizuyama (1990) were adopted for training and prediction curves of the velocity distributions using the five different machine learning techniques. Three theoretical formulas were employed for comparison and investigation, the power-law derived by Takahashi (1978) based on Bagnold dilatant flow, theory modified by Matsumura and Mizuyama (1990), and the two-region formula derived by Su et al. (1993). R-squared scores for each case were calculated to check the fitness of the machine learning results to the experimental data and then to verify the fitness of the theoretical formulas to the machine learning predictions. The quantified results revealed that machine learning schemes provide powerful approaches for building prediction models for velocity distribution of debris flows.