Developing speed‐related safety performance indicators from floating car data

Ambros, Jiří; Usami, Davide Shingo; Valentová, Veronika · 2022 · Crossref

DOI: 10.1049/itr2.12281

archive: archived pipeline: cataloged verified

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

Summary

This study addresses the need for proactive, non-crash-based safety performance indicators (SPIs) to monitor road safety at a macro-level. Traditional crash data is often considered reactive and incomplete, while existing surrogate measures of safety are typically limited to micro-level site assessments. The authors aimed to fill this gap by developing speed-related SPIs using nationwide floating car data (FCD) and validating their relationship to actual crash frequencies. The research specifically tested whether indicators derived from big data could serve as valid proxies for safety conditions across different road types. The methodology utilized FCD from approximately 700,000 vehicles in Italy, representing about 2% of the national fleet, alongside crash data from the CARE database. The authors calculated three candidate SPIs: the coefficient of variance (CV) for speed, the congestion index (CI), and the number of incidents (defined by high acceleration/deceleration values). Data was aggregated by province and road type (urban, motorway, rural). To validate these indicators, the study employed three methods: linear correlation analysis using Spearman coefficients, generalized linear models (negative binomial distribution) to predict crash frequency, and a ranking consistency test to compare SPI-based safety rankings against crash-based rankings. The results indicated that the number of incidents was the most effective SPI, particularly for motorways. Linear correlation analysis showed the highest coefficients between incidents and crashes (approximately 0.81 for motorways and 0.65 for urban roads), whereas speed-based indicators like CV and CI showed weaker correlations, especially on rural roads. The crash frequency models confirmed that incidents had the strongest explanatory power, with motorway models achieving the highest goodness-of-fit ($R^2$ values around 0.73–0.75). All explanatory variables, including vehicle-kilometers and speed metrics, demonstrated logical positive associations with crash frequency. The ranking consistency test further supported the validity of the incident-based indicator, showing high overlap with crash-based safety rankings. The significance of this work lies in demonstrating that FCD-derived incident counts can serve as robust, network-wide SPIs, particularly for high-speed infrastructure like motorways. While speed variability and congestion indices showed limited validity in this context, the incident metric provided a reliable proactive measure of safety. This finding supports the integration of big data into national safety monitoring frameworks, allowing for more timely and comprehensive safety assessments than traditional crash statistics alone. The study highlights the potential for using telematics data to identify unsafe road segments and evaluate interventions without waiting for crash occurrences.

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.

• incidence prevalence
• naturalistic crash near crash

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: crash risk outcomes
• Methodological Resource: dataset resource