Vehicle involvements in hydroplaning crashes: Applying interpretable machine learning
DOI: 10.1016/j.trip.2020.100176
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 challenge of identifying and analyzing hydroplaning crashes, which are often underreported in conventional crash databases due to the reliance on unstructured police narrative text. The primary research objective was to develop a framework using interpretable machine learning (IML) to classify crash narratives, specifically determining whether a hydroplaning incident involved a single vehicle or multiple vehicles. By extracting insights from textual data, the study aims to reveal underlying trends and contributing factors that are typically lost through manual interpretation of free-text reports. The researchers utilized seven years (2010–2016) of crash data from the Louisiana traffic crash database. Initial identification of hydroplaning crashes was performed using natural language processing (NLP) tools to search for keywords such as "hydroplane" and "hydroplaning," followed by word stemming. This process yielded 703 potential reports, which were manually verified by undergraduate students to remove false positives, resulting in a final dataset of 652 confirmed hydroplaning crashes. The study evaluated three machine learning algorithms: Support Vector Machine (SVM), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost). The data was split into training (67%), validation (13%), and test (17%) sets using stratified resampling. The framework included data preparation steps such as tokenization, stop-word removal, and the creation of domain-specific lexicons to enhance feature extraction from the narratives. The results indicated that the XGBoost model was the most effective classifier for distinguishing between single- and multiple-vehicle hydroplaning crashes. Descriptive statistics and chi-square tests revealed significant differences in crash attributes between the two groups. Multiple-vehicle hydroplaning crashes accounted for 32% of the total incidents and were more likely to occur on roadways with no abnormalities, under normal driver conditions, and on interstates. In contrast, single-vehicle crashes were more frequently associated with water on the roadway, inattentive drivers, and city streets. Vehicle condition analysis showed that engine failure was a prevalent factor in both groups, though slightly more common in multiple-vehicle crashes. The study also highlighted that while vehicle type and year did not show significant statistical differences between the groups, passenger cars were more prevalent in single-vehicle incidents. The significance of this research lies in its demonstration that interpretable machine learning can effectively process unstructured crash narratives to identify safety-critical patterns. By proving that quantitative modeling techniques can extract meaningful insights from textual data, the study provides a platform for reducing human error in crash surveillance and improving the understanding of hydroplaning mechanisms. The findings suggest that IML tools can serve as reliable decision-making aids for transportation engineers and safety analysts, enabling more precise identification of contributing factors and facilitating targeted interventions to mitigate hydroplaning risks.
Key finding
The eXtreme Gradient Boosting (XGBoost) model was the most effective classifier for determining the number of vehicle involvements in hydroplaning crashes based on police narrative text.
Methodology
dataset
Sample size: 652
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. Discovered via author_sweep_intake on 2026-05-28.
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | author_sweep | — | — | 2 | 2026-05-28 |
| archive | success | canonical_url | — | — | 11 | 2026-06-06 |
| extract | success | cached | — | — | 3 | 2026-06-10 |
| clean | success | clean | — | — | 1 | 2026-06-04 |
| chunk | success | chunk | — | — | 1 | 2026-06-04 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 1 | 2026-06-04 |
| enrich | success | — | — | — | 1 | 2026-05-28 |
| promote | success | — | — | — | 1 | 2026-06-04 |
| summarize | success | llm | qwen3.6-27b-prismaquant | summ-v5 | 2 | 2026-06-10 |
| tag | success | vector_similarity | — | — | 15 | 2026-06-11 |
| verify | success | — | — | — | 2 | 2026-06-10 |
Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.
Topics
Ranked by relevance to this paper. Hover a topic for its definition.