BEGIN:VCALENDAR
VERSION:2.0
PRODID:ECMLPKDD-MB
BEGIN:VEVENT
DTSTAMP;TZID=Europe/Dublin:20180826T200000
UID:_ecmlpkdd_77
DTSTART;TZID="Europe/Dublin":20180913T140000
DTEND;TZID="Europe/Dublin":20180913T142000
LOCATION:Nally
TRANSP:TRANSPARENT
SEQUENCE:1
DESCRIPTION:Large-scale real-time transport mode detection is an open challenge for smart transport research. We present the first method to detect transport modes taken by any traveling phone holder. We use anonymous Call Detail Records from the Greater Paris in collaboration with a mobile phone operator. We construct anonymized aggregated trajectories as sequences of mobile network locations, called sectors, where devices were detected. We use Bayesian inference to compute trajectories' transport modes probabilities. In this perspective, we engineer features using both mobile and transport networks and apply clustering on sectors in order to find transport probabilities given each visited sector. Using unsupervised evaluation metrics, we find 9 clusters best describe the region's transport usage. We construct 15% sectors labels to estimate clusters' probabilities. We derive prior distribution parameters from both trajectories and household travel survey. For model validation, we calculate daily average user trips at department scale. We find Pearson correlations with survey above 0.96 for road and rail modes, showing the model is performant and robust to sparse and noisy trajectories.
SUMMARY:Combining Bayesian Inference and Clustering for Transport Mode Detection from Sparse and Noisy Geolocation Data
CLASS:PUBLIC
END:VEVENT
END:VCALENDAR