Modeling Transit Patterns Via Mobile App Logs.

摘要

Transit planners need detailed information of the trips people take using public transit in order to design more optimal routes, address new construction projects, and address the constantly changing needs of a city and metro region. Better transit plans lead to better service and lower costs. Unfortunately, good rider origin-destination information is almost universally unavailable. In this project we have developed a new method for inferring rider origin-destination (O-D) trip stops in support of transit planning. The meteoric adoption of smartphones along with the growth of transit apps that provide vehicle arrival information at a stop generates a new data resource. Every time a user requests arrival information, the mobile service logs the user’s location, the time, and the specific stop they requested information about. Over time, a user’s request history functions as “bread crumbs” revealing where and when they have travelled. The goal of this project is to develop machine-learning models that can infer O-D for a transit service based on the request logs of individual users of mobile transit apps. This project builds on already deployed and extensively used Tiramisu app. In addition to the request log, Tiramisu data includes O-D trips recorded by users that we can use as ground truth for training the machine learning models. We will use this data to build a transit model that can derive results based on model phone app usage. Thus, we can produce models of transit use at a fraction of the cost. This approach also allows continuous O-D modeling, unlike traditional survey and sampling techniques. Note that, as far as we know, the Tiramisu app is a unique source of exact, large-scale, O-D information collected for research purposes. Other researchers have collected O-D using smartphones in small studies, but not through an extensively deployed app with over four years of historical data.