Predicting the traffic flow with using traffic sensors: An application for Istanbul

The place of transportation in human life is increasing day by day, and almost half of the society is spending one hour in traffic everyday [1]. According to Statistical Institute of Turkey, the number of motorized road vehicles registered in traffic in Istanbul reaches a number such as '3 651 166'. It has been tried to make estimations on traffic flow, to increase the efficiency of traffic operation, to give information about the way to go to traffic, and has attracted considerable attention as an application of intelligent transport systems [2]. Real time traffic data is collected from various sensors by the traffic directorates in Istanbul, as it is in many cities of the world. Variations such as the number of vehicles in the lines, the direction-based flow rate, the amount of occupation and the speed of the lines are typical universal data set variables [3]. The data set to be used in the study is taken from the IBB Traffic Directorate by means of a petition and an intention to carry out a scientific study and it follows a long route which focuses on the rather cramped roads of the city of Istanbul. The models which used to modelling traffic are ARIMA and Deep Multilayer Perceptron (DMLP). Data set separated in to parts as training and test sets to train and test models. The estimates on test set showed that DMLP is much more accurate than ARIMA for this data. DMLP is one of the deep learning algorithms. Deep learning models are famous with their ability to solve complex problems. After choosing DMLP as model, the aim of the study become predicting the traffic flow just with that model. In the study, it is determined which factors influenced the traffic flow by using artificial neural networks trained and then estimating the traffic flow of future periods.