Revealing route choice preferences through traffic-aware characterization and evaluation

A data-mining methodological framework to reveal drivers' route choice preferences from vehicle trajectory data through traffic-aware characterization and evaluation.

Citation info:

Wan, Z., & Dodge, S. (2025). Revealing drivers’ route choice preferences through traffic-aware characterization and evaluation. Journal of Location Based Services. (Accepted) http://dx.doi.org/10.1080/17489725.2025.2601133

Abstract

Understanding driver route choice behavior is essential for advancing navigation systems, autonomous driving, and intelligent transportation planning. While existing studies have identified a wide range of potential route choice factors across different regions, spatial and cultural heterogeneity may influence their relevance. To address this, we propose a traffic-aware, data-driven framework for revealing route choice preferences through interpretable modeling and clustering. The framework comprises three core components: (1) constructing a traffic-aware road network with hourly resolution from large-scale trajectory data, (2) identifying region-specific influential route choice factors by contrasting observed routes to randomized alternatives using a random forest model, and (3) clustering routes based on these factors to reveal distinct preference profiles. Applied to a dense vehicle trajectory dataset from Shenzhen, China, the framework uncovers three dominant route choice preferences: systematic time-focused routing, adaptive strategies prioritizing maneuver simplicity and traffic avoidance, and moderately efficient, habitual or routine-based behaviors. We further evaluate whether standard, non-preference-driven route generation algorithms can replicate these preferences and find that they fall short in capturing the full spectrum of driver behavior, particularly for context-sensitive and habitual routing styles. These findings underscore the value of the proposed framework in capturing meaningful behavioral diversity and point toward the need for preference-aware route generation strategies.

Descriptions

Note

Due to the file size limitation of GitHub, the relevant data is not included. Please refer to our Figshare project for the complete code with data.

Code

1. Preprocessing
   Run Preprocessing.ipynb in "Preprocessing" to preprocess the raw vehicle trajectory data. Run osrm_tracepoints.py and osrm_pts2edges.py in "Preprocessing/Map_Matching" to map match the trajectory to obtain routes.
2. Traffic-aware road network construction
   A traffic-aware road network with time-varying travel time information is used in this framework. Run Construct_Traffic-aware_Road_Net.py in "Traffic-aware_Road_Net" to construct it by estimating real travel time from vehicle trajectory data.
3. Synthetic route generation
   Run syn_routes_labeling.py, syn_routes_lkpen.py, syn_routes_lkelim.py, syn_routes_sim.py, and syn_routes_kshort.py in "Synthetic_Route_Generation" to generate synthetic routes using the labeling, link penalty, link elimination, simulation, and k-shortest-time paths algorithms, respectively.
4. Route characterization Run charact_routes.py in "Route_Characterization" to characterize the observed and generated routes.
5. Route choice factor identification
   Run Get_Factor_Dataset.ipynb and Random_Forest_Classifier.ipynb in "Significant_Choice_Factors" to identify the region-specific significant route choice factors by contrasting observed routes with randomly generated alternatives to identify and quantify the most influential route choice factors using a flexible and interpretable random forest model.
6. Preference-driven route choice set generation
   In "Preference_Choice_Sets", run Universal_Choice_Sets.ipynb to generate universal choice sets, Train_Pref_Clf.ipynb to train an MLP-based route choice preference classifier, and Pref_Choice_Set_Generation.ipynb to generate preference-aligned route choice sets. Finally, run Examine_Replication.ipynb to examine replication rates.