Enhancing the predictability of least-squares collocation through the integration with least-squares-support vector machine

Abstract

Least-squares collocation (LSC) is a crucial mathematical tool for solving many geodetic problems. It has the capability to adjust, filter, and predict unknown quantities that affect many geodetic applications. Hence, this study aims to enhance the predictability property of LSC through applying soft computing techniques in the stage of describing the covariance function. Soft computing techniques include the support vector machine (SVM), least-squares-support vector machine (LS-SVM), and artificial neural network (ANN). A real geodetic case study is used to predict a national geoid from the EGM2008 global geoid model in Egypt. A comparison study between parametric and soft computing techniques was performed to assess the LSC predictability accuracy. We found that the predictability accuracy increased when using soft computing techniques in the range of 10.2 %–27.7 % and 8.2 %–29.8 % based on the mean square error and the mean error terms, respectively, compared with the parametric models. The LS-SVM achieved the highest accuracy among the soft computing techniques. In addition, we found that the integration between the LS-SVM with LSC exhibits an accuracy of 20 % and 25 % higher than using LS-SVM independently as a predicting tool, based on the mean square error and mean error terms, respectively. Consequently, the LS-SVM integrated with LSC is recommended for enhanced predictability in geodetic applications.