DATA PREPROCESSING AND EMPIRICAL STUDY

DATA PREPROCESSING AND EMPIRICAL STUDY

We begin this section by introducing the taxi GPS tracescollected for our study. The large-scale taxi GPS data set wasacquired from approximately 7600 taxis served in Hangzhou, amegacity in China, for one year (April 2009 to March 2010).Each taxi is equipped with a GPS device that acquires the real-time taxi information, including its longitude/latitude, the timestamp, the passenger status (i.e., “occupied” or “vacant”), thedriving speed, and orientation. It uploads the data to a centralserver via a telecommunication network at a sampling rate ofonce per minute (i.e., the GPS sampling frequency is about onesample per minute). To avoid mistakes caused by device erroror network failure, we first filter out suspicious taxi records,such as taxi being vacant for the entire day, or lacking recordsfor extended periods of time, and obtain 6863 taxis to conductthe research.On a 2-D plane, a taxi’s moving trajectory over a timeinterval can be depicted by connecting the GPS points. Forinstance, Fig. 1 shows one taxi’s trajectory for a completepassenger-searching and passenger-delivery cycle, where redand blue lines correspond to the taxi’s vacant and occupiedstatuses, respectively. When the taxi is occupied, it is said tobe in the passenger-delivery stage; while the taxi is vacant, it isusually in the passenger-searching stage. The change from redto blue corresponds to a passenger pickup event, and the changefrom blue to red indicates a drop-off event.Before analyzing taxi driver’s service strategies, two datapreprocessing tasks need to be done.•Individual Taxi Driver GPS Trace Extraction: Almost allthe taxis in Hangzhou are served by two drivers, i.e., oneworks in daytime and the other in nighttime. As the shiftstatus of a taxi is not recorded in the taxi GPS traces, it isnecessary to detect when the shift handover takes place inorder to separate the taxi GPS traces of each driver. Onlywith each taxi driver’s GPS traces extracted, individualdrivers’ service strategies and revenues in different timeslots can be known.•Taxi Driver’s Performance Quantification: In order to dis-cover what service strategies are efficient and inefficient,we should be able to identify “good” and “ordinary”drivers by linking the service strategies with the revenuesgenerated. However, the taxi trip fares are not recordedin each taxi’s GPS traces; thus, it is necessary to quantifyeach taxi driver’s performance in each time slot accordingto the revenues calculated from the generated GPS traces.In the following sections, we will present how to deal withthe above two data preprocessing problems. We will also do anempirical study based on the processed data to visualize somestatistical results.
siva sakthi travels

For shared taxis, normally, the two drivers set up an agree-ment on when and where to hand over the taxi. Through interviews with taxi drivers in Hangzhou, we are told that theafternoon shift handover usually happens between 16:00 and18:00, and the morning shift handover takes place around06:00. The morning shift handover is generally more flexiblethan the one in the afternoon because many night shift driverspark the taxis at the agreed location around 03:00 so thatthe daytime drivers could take the taxis early if they wish.However, due to issues such as traffic, delay of last passenger deliveries, or personal reasons, it is hard to strictly obey the shifthandover agreement every day, particularly for the afternoonshift handover. Thus, the exact shift handover time detection becomes a nontrivial task in order to extract individual taxidriver’s GPS traces.https://srisivasakthitravels.com/