Use of artificial neural networks and support vector machines to predict lacking traffic survey data

The aim of this paper was to predict lacking data from a traffic survey along a principal highway in Bangladesh using artificial neural network (ANN) combined with the support vector machine (SVM). Traffic data were obtained at an hourly rate using a methodical inquiry over a four-year period at the Jamuna toll collection point, which is located along the North Bengal corridor of Bangladesh. Two evolutionary computational statistical procedures were used along with its corresponding numerical model. The neural network and SVM were fed with data from 13 recurring weeks over a four-year period. The missing data were predicted with significant accuracy using both methods. Accuracy of the methods was compared, which showed that the SVM method is much more accurate than the ANN technique. Combination of both the ANN and SVM models can be used to obtain trends in traffic data more accurately.