Advanced OR and AI Methods in Transportation CALIBRATION OF LOGIT MODAL SPLIT MODELS WITH F.pdf

Advanced OR and AI Methods in Transportation CALIBRATION OF LOGIT MODAL SPLIT MODELS WITH FEED FORWARD BACK-PROPAGATION NEURAL NETWORKS Hilmi Berk CELIKOGLU1 Abstract. The presented study examines the possibilities of obtaining better logit mode choice models for home-based work trip purpose in Istanbul metropolitan area by calibrating binary logit modal split models with the employment of feed forward back-propagation algorithm trained neural networks. A two-variable logit model with the trip cost and the trip time variables is calibrated to split trips to private car and public transport modes. The calibration data is aggregated at an appropriate level considering the previous studies’ and master plans’ outcomes. Following the neural network calibrations, the two-variable logit model is calibrated with linear regression method. The results were then compared. 1. Introduction Istanbul, the largest urban settlement area in Turkey, has enormous commercial, cultural, and historical significance. This study examines the possibilities of obtaining better logit mode choice models for daily home-based work trips in Istanbul metropolitan area by calibrating binary logit modal split models with the employment of feed forward back- propagation algorithm trained neural networks. By means of the intersectorial trip data obtained by the aggregation process at a level of 22 sectors; a two-variable binary logit model with trip cost and trip time variables is calibrated for home-based work trips to split trips to private car and public transport modes. Data aggregation for the calibration is done considering the outcomes of previous studies and master plans for Istanbul metropolitan area. Following the neural network calibrations, the two-variable binary logit model is calibrated with linear regression method. The results are then compared in terms of selected performance criteria. 1 Technical Univer