Scalable Unsupervised Multi-Criteria Trajectory Segmentation and Driving Preference Mining

TL;DR

This paper introduces scalable techniques for analyzing large trajectory datasets to understand driving preferences and identify key points, enabling personalized routing and trajectory explanation.

Contribution

It presents novel, scalable methods for trajectory segmentation and driving preference mining that work efficiently on massive datasets.

Findings

Effective via-point identification in large datasets

Successful recovery of individual driving preferences

Techniques are highly parallelizable and scalable

Abstract

We present analysis techniques for large trajectory data sets that aim to provide a semantic understanding of trajectories reaching beyond them being point sequences in time and space. The presented techniques use a driving preference model w.r.t. road segment traversal costs, e.g., travel time and distance, to analyze and explain trajectories. In particular, we present trajectory mining techniques that can (a) find interesting points within a trajectory indicating, e.g., a via-point, and (b) recover the driving preferences of a driver based on their chosen trajectory. We evaluate our techniques on the tasks of via-point identification and personalized routing using a data set of more than 1 million vehicle trajectories collected throughout Denmark during a 3-year period. Our techniques can be implemented efficiently and are highly parallelizable, allowing them to scale to millions or billions of trajectories.