Driving Behavior Identification and Real-World Fuel Consumption Estimation With Crowdsensing Data

Pirayre, Aurélie; Michel, Pierre; Rodriguez, Sol Selene; Chasse, Alexandre · 2022 · OpenAlex-citations

DOI: 10.1109/tits.2022.3169534

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the significant discrepancy between official laboratory-measured fuel consumption and real-world usage, which can reach 30–40%. The authors aim to estimate real-world fuel consumption by identifying driving behaviors using crowdsensing data, specifically GPS data from smartphones, without requiring direct fuel consumption measurements or proprietary vehicle data. The motivation is to provide a posteriori estimation method that relies on semantic information (such as origin-destination and macro-behavior) rather than recorded sensor traces for new drivers. The methodology utilizes GPS data (velocity, acceleration, longitude, latitude) collected via the IFPEN’s GECO AIR application from over 73,000 trips in France. Trips are segmented into four road types (30 km/h zones, urban, extra-urban, highway) using map-matched speed limit data from HERE. For each segment, 18 statistical features are derived, including velocity metrics (average, max, null speed percentage), acceleration metrics (average, max, sparsity indices), and characteristics of the 2D velocity-acceleration probability density. An unsupervised K-prototype classification algorithm is applied to these mixed numerical and categorical features to identify distinct driving behaviors for each road type. Expert interpretation assigns semantic labels (e.g., dynamic, slow, traffic jam) to the resulting clusters. Representative velocity profiles are then generated for each behavior using Markov chains. These profiles are aggregated to reconstruct realistic driving cycles, which are input into a quasi-static fuel consumption model to estimate real-world fuel use. The results demonstrate that the unsupervised classification successfully identifies distinct driving behaviors linked to specific road conditions and driver styles. Statistical analysis confirms that the identified clusters exhibit discriminant fuel consumption levels. The proposed estimation method achieves promising performance, with more than half of the recorded trips estimated with less than 10% error compared to actual fuel consumption. The approach effectively captures the impact of trip-related factors (road type, traffic) and driver-related factors (aggressiveness, speed) on fuel usage. The significance of this work lies in its ability to estimate real-world fuel consumption a posteriori using only semantic input data, such as questionnaire responses, without needing continuous GPS records or manufacturer-specific fuel data. This facilitates the assessment of realistic usage-based fuel consumption for new drivers and supports applications in eco-driving recommendations, sustainable transport planning, and autonomous vehicle development. By relying on accessible smartphone GPS data and unsupervised learning, the method offers a scalable solution for understanding the complex interplay between driving behavior and environmental impact.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• exposure measurement

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Empirical Findings: observational prevalence
• Theoretical Contribution: computational model