Applying Mambaattention, Tabpfn, and Tabtransformers to Classify Sae Automation Levels in Crashes
DOI: 10.2139/ssrn.5229751
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 critical need for accurate classification of Society of Automotive Engineers (SAE) automation levels in motor vehicle crashes, a task complicated by the increasing prevalence of automated vehicles (AVs) and the limitations of traditional rule-based or statistical methods. Existing approaches often fail to capture the nuanced dynamics of human-machine interaction or distinguish between different automation tiers, leading to potential errors in crash causality analysis and policy development. To bridge this gap, the authors evaluate three advanced tabular deep learning models—MambaAttention, TabPFN, and TabTransformer—on their ability to classify crashes into SAE Level 1 (Assisted Driving), Level 2 (Partial Automation), and Levels 3–5 (Advanced Automation). The research utilizes structured crash data from Texas in 2024, comprising 4,649 initial cases. To address class imbalance, the dataset was balanced using the SMOTEENN technique, resulting in a unified training and evaluation dataset of 7,300 records. The study compares the performance of MambaAttention, which employs a state-space perspective for linear-time sequence modeling; TabPFN, a transformer-based foundation model capable of zero-shot inference without task-specific training; and TabTransformer, which uses contextual embeddings for categorical features. The experimental design focuses on assessing classification accuracy, computational efficiency, and robustness across the different automation categories. The results demonstrate that MambaAttention achieved the highest overall performance, with F1-scores of 88% for SAE Level 1, 97% for SAE Level 2, and 99% for SAE Levels 3–5. TabPFN exhibited strong robustness, particularly in zero-shot inference scenarios and for rare crash categories, highlighting its utility in settings with limited data. In contrast, TabTransformer underperformed significantly, achieving an F1-score of only 55% for Partial Automation (Level 2) crashes. This poor performance suggests that TabTransformer struggles to model the complex shared control dynamics inherent in mid-level automation systems. These findings underscore the potential of specialized tabular deep learning architectures to enhance the accuracy of AV crash analysis. The superior performance of MambaAttention and the zero-shot capabilities of TabPFN indicate that integrating such models into crash investigation frameworks can improve the identification of automation-specific risk factors. This capability is vital for developing targeted safety regulations, evaluating AV system accountability, and designing mitigation strategies for high-risk conditions associated with mid- and high-level automation technologies.
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 2026-05-27.
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | author_sweep | — | — | 2 | 2026-05-27 |
| archive | success | canonical_url | — | — | 5 | 2026-06-09 |
| extract | success | cached | — | — | 2 | 2026-06-10 |
| clean | success | clean | — | — | 1 | 2026-06-09 |
| chunk | success | chunk | — | — | 1 | 2026-06-09 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 1 | 2026-06-09 |
| enrich | failed | — | — | — | 4 | 2026-07-02 |
| promote | success | — | — | — | 1 | 2026-06-04 |
| summarize | success | llm | qwen3.6-27b-prismaquant | summ-v5 | 1 | 2026-06-10 |
| tag | success | vector_similarity | — | — | 8 | 2026-06-11 |
| verify | success | — | — | — | 1 | 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.
- crash typology
- naturalistic crash near crash
- situational awareness
- pre crash contributing factors
- driverless ads
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
- Theoretical Contribution: conceptual framework