Inferring human mobility patterns from taxicab location traces

Abstract

Taxicabs equipped with real-time location sensing devices are increasingly becoming popular. Such location traces are a rich source of information and can be used for congestion pricing, taxicab placement, and improved city planning. An important problem to enable these application is to identify human mobility patterns from the taxicab traces, which translates to being able to identify pickup and dropoff points for a particular trip. In this paper, we show that while past approaches are effective in detecting hotspots using location traces, they are largely ineffective in identifying trips (pairs of pickup and dropoff points). We propose the use of a graph theory concept -stretch factor in a novel manner to identify trip(s) made by a taxicab and show that a Hidden Markov Model based algorithm can identify trips (using real datasets from taxicab deployments in Shanghai and partially simulated datasets from Stockholm) with precision and recall of 90-94% −, a significant improvement over past approaches that result in a precision and recall of about 50-60%.

Exploring human mobility with multi-source data at extremely large metropolitan scales

“…Transit Data: Transit GPS data are another important source for research in human mobility, e.g., identifying human mobility based on data from taxicabs [7], buses [3], subways [12], and private cars [8]. In contrast, our method is based on data from the entire set of urban transit networks correlated with data from cellular networks, instead of sampling residents using a specific transit mode.…”

“…In this work, we focus on the mobility pattern inference, instead of dividing mobile traces. Therefore, we utilize one of the state-of-the-art methods [7] to obtain the trips based on the trace. In short, this method utilizes a graph theory concept called stretch factor to find several anchor points on a continuous trace as alternative origins and destinations, thus identifying the trips on the trace.…”

“…The inference about G c and G t is a classic topic for many data mining applications [7]. To focus on our key novelty, we maintain and update the conditional probability distributions over the set of destinations that a particular cellphone or transit user can go to, given the previous trip history and three real-time contexts, i.e., the current location, the time of day and the day of week.…”

“…Based on various empirical datasets, several human mobility models have been proposed [7] [9] [10] [11] [15] [17] that capture human mobility to a certain degree yet all share the common drawback of biased sampling:…”

“…• The research based on urban transit data assumes that residents of large urban cities usually employ urban transit (i.e., taxicabs [7], buses [3] or subway [12]) for travel. But based on our empirical results in Section 3, an approach based on data from one type of transit has a bias against residents using other types of transit.…”