Driving Big Data: A First Look at Driving Behavior via a Large-Scale Private Car Dataset

摘要

The increasing number of privately owned vehicles in large metropolitan cities has contributed to traffic congestion, increased energy waste, raised CO2 emissions, and impacted our living conditions negatively. Analysis of data representing citizens' driving behavior can provide insights to reverse these conditions. This article presents a large-scale driving status and trajectory dataset consisting of 426,992,602 records collected from 68,069 vehicles over a month. From the dataset, we analyze the driving behavior and produce random distributions of trip duration and millage to characterize car trips. We have found that a private car has more than 17% probability to make four trips per day, and a trip has more than 25% probability to last 20-30 minutes and 33% probability to travel 10 Kilometers during the trip. The collective distributions of trip mileage and duration follow Weibull distribution, whereas the hourly trips follow the well known diurnal pattern and so the hourly fuel efficiency. Based on these findings, we have developed an application which recommends the drivers to find the nearby gas stations and possible favorite places from past trips. We further highlight that our dataset can be applied for developing dynamic Green maps for fuel-efficient routing, modeling efficient Vehicle-to-Vehicle (V2V) communications, verifying existing V2V protocols, and understanding user behavior in driving their private cars.