Investigating mandatory and non-mandatory trip patterns based on socioeconomic characteristics and traffic analysis zone features using deep neural networks

Abstract

Forecasting travel demand is a classic problem in transportation planning. The models made for this purpose take the socioeconomic characteristics of a subset of a population to estimate the total demand, mainly using random utility models. However, with machine learning algorithms fast becoming key instruments in many transportation applications, the past decade has seen the rapid development of such models for travel demand forecasting. As these algorithms are independent of assumptions, have high pattern recognition ability, and often offer promising results, they can be effective alternatives to discrete choice models for forecasting trip patterns. This paper aimed to predict mandatory and non-mandatory trip patterns using a Deep Neural Network (DNN) algorithm. A dataset containing Metropolitan Washington Council of Government Transportation Planning Board (MWCGTPB) 2007–2008 survey data and a dataset containing traffic analysis zones’ characteristics (TAZ) were prepared to extract and predict these patterns. After the modeling phase, the models were evaluated based on accuracy and Cohen’s kappa coefficient. The estimates of mandatory and non-mandatory trips were found to have an accuracy of 70.87% and 50.02%, respectively. The results showed that a DNN could find the relationship between socioeconomic factors and trip patterns. This can be helpful for transportation planners when they are trying to predict travel demand.