Interpolation Techniques for Predicting the Movement of Cyclists

Abstract

Listen

Translate article

Planning the development of cycling infrastructure underlines the importance of predicting the movement of cyclists inside cities. The main goal of the paper is to examine different interpolation techniques in order to estimate the most accurate predicts. The research was conducted with primarily collected data on pre-selected intersections. The data set included 30 input points measured during cyclists’ rush hour in the morning, 5:30–9:00 a.m., and in the afternoon, 2:00–5:00 p.m. To choose an appropriate interpolation method IDW, EBK, RBF, Ordinary Kriging (OK) with spherical variogram, Ordinary Kriging (OK) with linear variogram, Simple Kriging (SK) with spherical variogram and Simple Kriging (SK) with linear variogram) were considered based on state of the art analysis and further examined. For selecting a suitable interpolation technique, the cross validation was taken into account comparing Mean Error (ME), Root Mean Square Error (RMSE), Root Mean Square Standardized Error (RMSSE) and Average Standard Error (ASE). The cross-validation showed that IDW and RBF have worst results although IDW was the most accurate in prediction of furthest point. Opposite, EBK and OK (spherical variogram) achieved very similar values bringing best predicts. Though kriging is very accurate interpolator, the behaviour of cyclists is determined by many other factors which can not be completely included during kriging.