Authors:
Carlos Carvalho
1
;
Moisés Ramires
2
and
Rui José
1
Affiliations:
1
Centro Algoritmi/LASI, University of Minho, Guimarães, Portugal
;
2
CCG, University of Minho, Guimarães, Portugal
Keyword(s):
OpenStreetMap, Map-Matching, Micro-Mobility, Urban Cycling.
Abstract:
Road network data models are a key element for many cycling services. However, cyclists often ride unconventional paths that may not be properly represented in those models. This may cause various types of map-matching anomalies, where the map-matched route does not correspond to the real route. In this work, we assess a set of classification models to automatically detect and classify these map-matching anomalies. Using OpenStreetMap road network, we generated the map-matched routes for a dataset of 98 cycling GPS traces. To produce ground-truth data, we visually inspected each result to identify and classify every map-matching anomaly, and computed several similarity measures between each GPS trace and the respective map-matched segment. Based on this data, we trained several classification models with different feature engineering approaches to perform binary and multi-class classification. The results show that binary classifiers can be very effective in the identification of map
-matching anomalies. The best model, a XGBoost classifier, obtained an F1 Score of 0.906 and an accuracy of 0.893, which outperform other methods. However, the multi-class classifiers had lower performance. This ability to automatically detect and classify map-matching anomalies may help to systematically improve road network models and consequently improve information provided to cyclists and decision-makers.
(More)