Artificial Intelligence Approach To Coordinate Transformation Using Group Method Of Data Handling And Least Square Support Vector Machine

Abstract

Transforming coordinates from the WGS84 datum to the Accra 1929 reference frame is still a problem for geospatial professionals in Ghana, as long as Ghana has not moved to the geocentric datum. This research applied the Bursa-Wolf transformation model to determine transformation parameters, changed the World Geodetic System 1984 (WGS84) points to the Accra datum and then projected the transformed coordinates onto the Transverse Mercator 1o NW projection system being used in Ghana. The Bursa-Wolf results were compared to the Group Method of Data Handling (GMDH) and the Least Square Support Vector Machine (LS-SVM). For all methods, the geodetic coordinates of the global system were transformed to the geodetic coordinates of the War Office before projection. To get the optimal performance of the model, the K-fold cross-validation approach was applied. Thirty-four points were used and divided into five folds. The average of the horizontal positional error of the five-folds gave the model the best output. The result reveals that the LS-SVM, GMDH and the Bursa-Wolf models average root mean square horizontal positional error of 2.5541, 3.0312, 3.3396, respectively; average mean horizontal positional error of 2.1904, 2.6191, 2.7578; and averaged standard deviations of 1.7, 2.139, 2.3563 respectively. It shows that LS-SVM is of better performance than the GMDH and the Bursa-Wolf model. Since the horizontal error of the LS-SVM is higher than the allowable ±0.9114 m standard for horizontal measurement in Ghana, the LS-SVM was recommended to be used for only low-order surveys like data collection for GIS databases, and topographic surveys