Tide modelling using support vector machine regression
Author(s)
Okwuashi, Onuwa
Ndehedehe, Christopher
Griffith University Author(s)
Year published
2017
Metadata
Show full item recordAbstract
This research explores the novel use of support vector machine regression (SVMR) as an alternative model to the conventional least squares (LS) model for predicting tide levels. This work is based on seven harmonic constituents: M2, S2, N2, K2, K1, O1 and P1. The SVMR is modelled with four kernel functions: linear, polynomial, Gaussian radial basis function and neural. The computed r-square and root mean square error for the linear, polynomial, Gaussian radial basis function and neural SVMR kernels as well the LS indicate a strong correlation between the observed and predicted tides. But for the linear kernel the results of ...
View more >This research explores the novel use of support vector machine regression (SVMR) as an alternative model to the conventional least squares (LS) model for predicting tide levels. This work is based on seven harmonic constituents: M2, S2, N2, K2, K1, O1 and P1. The SVMR is modelled with four kernel functions: linear, polynomial, Gaussian radial basis function and neural. The computed r-square and root mean square error for the linear, polynomial, Gaussian radial basis function and neural SVMR kernels as well the LS indicate a strong correlation between the observed and predicted tides. But for the linear kernel the results of all the kernels are slightly better than the LS. The statistical tests of the difference between the observed tide and the LS and SVMR predicted tides and between the LS and SVMR predicted tides are insignificant at the 95% confidence level.
View less >
View more >This research explores the novel use of support vector machine regression (SVMR) as an alternative model to the conventional least squares (LS) model for predicting tide levels. This work is based on seven harmonic constituents: M2, S2, N2, K2, K1, O1 and P1. The SVMR is modelled with four kernel functions: linear, polynomial, Gaussian radial basis function and neural. The computed r-square and root mean square error for the linear, polynomial, Gaussian radial basis function and neural SVMR kernels as well the LS indicate a strong correlation between the observed and predicted tides. But for the linear kernel the results of all the kernels are slightly better than the LS. The statistical tests of the difference between the observed tide and the LS and SVMR predicted tides and between the LS and SVMR predicted tides are insignificant at the 95% confidence level.
View less >
Journal Title
Journal of Spatial Science
Volume
62
Issue
1
Subject
Physical geography and environmental geoscience
Geomatic engineering
Geospatial information systems and geospatial data modelling