MARBLE: A Multi-Agent Rule-Based LLM Reasoning Engine for Accident Severity Prediction

Jiashu Zhang; Kaleem Ullah Qasim

REVIEW 5 major objections 7 minor 45 references

MARBLE: A Multi-Agent Rule-Based LLM Reasoning Engine for Accident Severity Prediction

T0 review · 5 major / 7 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read MARBLE claims that splitting accident-severity prediction across a team of specialized small-model agents reaches nearly 90% accuracy on UK and US crash data, far above monolithic machine learning and prompting baselines.

desk verdict The architecture is novel and the ablation is useful, but the headline accuracy claim is confounded by an unfair baseline comparison and internal inconsistencies. read the letter →

arxiv 2507.04893 v1 pith:S3RH3ESH submitted 2025-07-07 cs.AI cs.CLcs.MA

Kaleem Ullah Qasim , Jiashu Zhang This is my paper · ORCID

classification cs.AIcs.CLcs.MA

keywords accidentseveritypredictionmulti-agentreasoningsmalllanguagemodelsrule-basedcoordinationclassimbalancestructuredtabulardatainterpretableAILLMprompting

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

MARBLE is a system for predicting traffic accident severity that splits the task across a team of small, specialized reasoning agents: one conventional machine-learning model sees all features, while separate small-language-model agents each reason over one semantic slice of the data, such as weather and visibility, road geometry, or time of day. A deterministic rule-based coordinator then fuses the agents' votes with weights that deliberately favor rare, high-severity outcomes. The paper claims this design reaches 89.5% and 89.8% accuracy on UK and US crash datasets, whereas classical machine-learning baselines plateau between 41% and 47% and prompting-only methods stay below 32%. The claim matters because real accident data is noisy and severely imbalanced, and the authors argue that modular feature decomposition plus structured fusion, rather than a larger model or a longer prompt, is what unlocks the gain. If the claim holds, the same architecture offers a generalizable and interpretable recipe for structured prediction in safety-critical settings.

What carries the argument

The load-bearing mechanism is the feature-space decomposition with structured fusion. Feature projections $\pi_a$ give each agent a coherent subset ($\mathcal{F}_a \subset \mathcal{F}$), so the environmental, locational, spatial, and temporal agents each reason over only their slice of the input while the ML agent sees everything. The coordinator $\Phi_{\mathrm{RB}}$ then computes a weighted vote $S'_k = \sum_{a \in A} w_a\, c_a\, \beta_k\, \mathbb{I}(\hat{y}_a = k)$ using static agent weights $w_a = (3.0, 1.5, 1.2, 1.0, 1.0)$ and class-importance factors $\beta_k = 1.2$ for the rare classes $\{1,4\}$ and $1.0$ otherwise, with an override that trusts the ML agent when its confidence exceeds $0.75$ and at least one corroborating SLM agrees. SLM confidences are recalibrated with a heuristic boost that adds up to $0.1$ to high-confidence rare-class predictions, and an agreement-based boost $\Delta C$ raises the final confidence when multiple domain agents concur. The final decision layer $F_{\mathrm{final}}$ picks between the ML output and the coordinator using class-dependent confidence thresholds ($0.4$ for rare classes, $0.5$ otherwise) and a tie-breaking weight that again favors rare classes. This explicit rarity-aware machinery is what the paper credits for converting several mediocre local predictions into a high-confidence global decision.

What would settle it

Run a matched-budget experiment: train the Random Forest and Gradient Boosting baselines on the same number of training instances MARBLE's ML agent actually used, and retrain that ML agent on the 500-per-class budget, testing all variants on the same held-out split with per-class recall reported.

Watch

Extended reading notes

Core claim

The paper's central claim is that a hybrid multi-agent ensemble — a machine-learning agent plus several domain-specialized small-language-model agents coordinated by hand-designed rules — can predict accident severity in four classes at nearly 90% accuracy on two national datasets, outperforming the best traditional classifier by more than 42 percentage points and every tested monolithic prompting strategy by more than 58 points. The authors attribute the gain to the decomposition itself: each agent receives only a semantically coherent subset of features, which avoids the context saturation and entangled reasoning that plague single-prompt LLM approaches to tabular data, while the rule-based coordinator exploits agent confidences, rarity-aware class weights, and an ML-override rule to settle disagreements. The authors further report that the best configuration runs on a 1.7B-parameter model, needs no data balancing or augmentation, keeps per-agent reasoning traces for interpretability, and stays above 87% accuracy under simulated extreme class-imbalance shifts — evidence they take to show the framework sets a new practical ceiling for this task under real-world noise.

Load-bearing premise

The claimed 40-point advantage rests on every method being given a fair comparison, yet the ML baselines were trained on only 500 samples per severity class while the paper does not state how many samples trained MARBLE's own ML agent, so the gap may partly reflect differing data budgets rather than the multi-agent architecture.

Editorial extensions

If this is right

On both the UK and US datasets, MARBLE reports accuracy of 89.5% and 89.8% with macro F1 around 0.90, more than 40 points above the best classical baseline and more than 55 points above the best prompting method tested.
Rule-based coordination is reported to beat LLM-based coordination by up to 23.7 points over the same agent set, so the structured fusion logic, not the coordinator's language model, carries most of the gain.
Ablation removes each agent in turn: dropping the environmental agent costs the largest relative accuracy drop (roughly 52% on the UK set and 75% on the US set), showing the result depends on the interplay of all agents with the coordinator rather than any single component.
Under six simulated class-imbalance regimes, accuracy stays above 87% with rule-based coordination and degrades far less than the LLM-coordinated variant, indicating resilience to skewed label distributions without resampling or augmentation.
Because the best result uses a 1.7B-parameter model with structured traces of every agent's reasoning, the framework is presented as deployable on consumer hardware while remaining interpretable.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A reader should not take the headline margin at face value until the training budgets are matched: the paper states the classical baselines saw only 500 samples per class but never states the training-set size for its own ML agent, so a fair retest could shrink the reported gap considerably.
The design pattern — slice tabular features by meaning, reason per slice with small models, fuse with explicit rarity-aware weights — is not specific to traffic accidents and should transfer to other imbalanced structured-prediction tasks such as fraud detection, insurance claims, or clinical triage, where recall on rare outcomes matters more than overall accuracy.
A natural follow-up the paper leaves implicit is learning the agent weights $w_a$ and boost parameters from validation data instead of fixing them by hand; if learned weights matched or beat the hand-set ones, the architecture would become fully adaptive.
The confidence-boost heuristics for rare classes are a form of recalibration; replacing them with proper per-agent probability calibration (for example, isotonic regression on validation outputs) would give the reported confidence scores a meaning that can be compared across deployment settings.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Colleague—

The headline result isn't established by the experiments as written. MARBLE reports 89.5/89.8% accuracy on the UK/US datasets, but the comparison against traditional ML baselines is apples-to-oranges: the baselines were trained on 500 samples per class, while the appendix indicates the integrated ML agent is trained on the full 150k/180k subsets. Without a matched training budget or the ML agent's standalone accuracy, the 42-point gap could be a data-size artifact rather than evidence for the multi-agent architecture.

That said, the architecture is worth a look. Decomposing the feature space across small language model agents, each with its own prompt, and fusing via a weighted rule-based coordinator is a reasonable idea, and I don't recall it being applied to tabular accident severity prediction before. The ablation in Fig. 2 is informative—removing the environmental agent collapses accuracy, which is a concrete and useful finding. The internal comparison between rule-based and LLM-based coordination (Table II) is also a fairer test and shows the rule-based version is sturdier.

The soft spots are real. Besides the baseline mismatch, the paper contradicts itself: Table II's footnote says Smollm2-1.7B was fine-tuned for agent-specific usage, while RQ1 claims results come 'without any large-scale pretraining or dataset-specific tuning.' And RQ3 reports the complete MARBLE at 87.2%/89.5%, which differs from Table I's 89.5%/89.8%; the paper doesn't explain which number is correct. Then there are the many hand-fitted parameters (agent weights, class importance factors, confidence thresholds, override conditions) all derived from validation data, with no sensitivity analysis. No code or data are released, so the key numbers can't be checked. That's a lot of tuning for a system whose main selling point is interpretability and rule-based transparency.

None of this kills the underlying idea. A cleaner experiment with matched baselines, the ML agent's standalone accuracy reported, and a sweep over the coordination constants would tell us whether the architecture adds real value. As it stands, the evidence is not persuasive.

I'd still send this to peer review—the architecture is novel enough and the domain important enough to deserve referee time—but the reviews should ask for a corrected comparison, not incremental tweaks.

Summary: worth engaging with for the design and the ablation, not for the headline numbers.

Referee Report

5 major / 7 minor

Summary. The paper proposes MARBLE, a multi-agent system for accident severity prediction that combines one machine-learning agent (RandomForest/GradientBoosting-style classifier) with four domain-specialized small-language-model agents (environmental, temporal, infrastructural, spatial/dynamic) and a coordinator. The coordinator is either a rule-based weighted-voting mechanism with hand-set thresholds and class-importance factors, or an LLM-based meta-reasoner. The authors evaluate on UK STATS19 and US accident subsets and report 89.5% and 89.8% accuracy, which they compare to traditional ML baselines (40-47%) and prompting baselines (25-32%). They also report ablations, coordination-strategy comparisons, and class-imbalance simulations. The central claim is that feature-decomposed multi-agent SLM reasoning, rather than data volume or model scale, drives the dramatic improvement.

Significance. If the headline result were established with matched experimental budgets and proper validation, it would be a noteworthy demonstration that lightweight, modular multi-agent SLM reasoning can beat monolithic tabular ML and monolithic prompting on a strongly imbalanced structured-prediction task. The paper has genuine strengths: the feature-space decomposition is interpretable by design, the architecture keeps structured traces of agent reasoning, and the authors include ablation studies (Figs. 2, 6) and robustness experiments under simulated imbalance. However, the current manuscript does not establish the headline claim: the comparison conditions are not matched across systems, several internal contradictions remain unresolved, and the many hand-fitted coordination constants raise a real overfitting risk. No code, data splits, or fine-tuning details are provided, so the reported numbers cannot be reproduced or independently checked.

major comments (5)

[V (RQ1) and Appendix; III-C1] The comparison between MARBLE and the traditional ML baselines is not matched in training data. Section V (RQ1) states that each ML/DL baseline was trained on only 500 samples per severity class, while the Appendix states that the full 150,000-instance US subset and 180,000-instance UK subset are used 'for training its integrated machine learning component.' Section III-C1 describes the ML agent as a RandomForest/GradientBoosting classifier optimized on Dtrain. The training set size for MARBLE's own ML agent is never explicitly reported, nor is the ML agent's standalone accuracy. Since the ML agent is a core component of MARBLE, the reported 42-point improvement could be entirely due to data-volume differences rather than the multi-agent architecture. Please re-run all baselines under matched training budgets (including the same class-imbalance handling) and report the ML agent's standalone performance.
[Table II footnote; V (RQ1)] There is a direct contradiction about fine-tuning. RQ1 claims the best result is achieved 'without any large-scale pretraining or dataset-specific tuning,' but Table II's footnote states that 'HuggingFace's SmollM2-1.7B was fine-tuned for agent-specific usage.' This matters because the fine-tuned Smollm2-1.7B is the model behind the best reported numbers (89.5%/89.8%). Clarify exactly which components were fine-tuned, on what data, and how this affects the 'no dataset-specific tuning' claim.
[V (RQ3) vs Table I] The reported accuracy for the complete MARBLE system is inconsistent across the paper. RQ3 states that the complete MARBLE configuration yields 87.2% (UK) and 89.5% (US) accuracy, whereas Table I reports 89.5% (UK) and 89.8% (US) for the same best configuration. The UK discrepancy is 2.3 percentage points, which is too large to ignore. The authors must explain which number is correct and reconcile the discrepancy, as this directly affects the headline claim.
[III-E and III-F] The final accuracy depends on a large number of constants that are described as 'derived empirically from validation data' (Section III-E1): agent importance weights w (Eq. 14), class importance factors beta (Eq. 15), SLM calibration boosts (Eq. 10), ML override thresholds tau_ML_high and tau_ML_corrob, agreement boost Delta_C (Eq. 19), and final decision thresholds and tie-break weights (Eq. 24). The paper does not describe the validation procedure, the size of the validation set, or how these constants were selected. With this many free parameters, the near-90% accuracy could reflect overfitting to the evaluation data. Please provide a nested validation scheme and a systematic sensitivity analysis; Fig. 6 only partially addresses this.
[V; Tables I and II] The experimental reporting is incomplete. Table I says the reported metrics are averaged over 5 cross-validation runs, but no standard deviations, confidence intervals, or significance tests are provided. The paper also does not describe the train/test split, the class distribution, or the number of instances per split. Without these, the reader cannot assess whether the very large reported gaps are stable across splits or statistically meaningful.

minor comments (7)

[III-B] The text references 'Appendix III' but the appendix is not numbered; please correct the cross-reference.
[V] The opening paragraph of Section V says the framework leverages 'LLaMA 3.2 3B Instruct,' while the best result uses Smollm2-1.7B (per Table II). This inconsistency should be resolved.
[Fig. 4 caption] The caption refers to 'analysis 3' without defining it; please make the reference self-contained.
[Fig. 2] Panel B reports relative accuracy drops of 52.4% (UK) and 74.8% (US) when the environmental agent is removed; given overall accuracies around 87-89%, these percentage drops should be contextualized as relative drops to avoid confusion with absolute accuracy changes.
[III-F, Eq. (25)] The variable Coverride is used in Eq. (25) but is never formally defined; the rule-based coordination section describes an override condition in prose, but the final decision logic needs an explicit definition.
[VII (Limitations)] The Limitations section candidly mentions SLM stability, delay, and centralized coordination, but it does not acknowledge the unmatched training budgets between MARBLE's ML agent and the ML baselines, which is the most serious threat to the validity of the headline comparison.
[Abstract] The phrase 'redefines the practical ceiling' is an overstatement given the unresolved comparison and validation issues; please temper the language.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation found; the headline accuracy is an empirical cross-validated measurement, not a by-construction consequence of the method's equations. The central comparison is, however, confounded by unequal training budgets, which is a fairness/correctness issue rather than circularity.

full rationale

The reported 89.5%/89.8% accuracies are empirical measurements on held-out folds (Table I note: 'averaged metrics from 5 independent runs in a cross-validation framework'), and no equation in Sections III-E or III-F computes test accuracy from the hand-set weights, beta factors, calibration boosts, or thresholds. The only self-citation, reference [18], supports a general statement about decompositional power and is not load-bearing. There is therefore no step in the derivation chain that reduces to its own input by construction. That said, three non-circular but serious internal-consistency problems are present: (1) Section V RQ1 trains ML baselines on 'only 500 samples per severity class' while the Appendix states that the full 150,000/180,000-instance datasets are used 'for training its integrated machine learning component,' so the 42-point advantage over baselines conflates architecture with training data volume; (2) Table II's footnote that Smollm2-1.7B 'was fine-tuned for agent-specific usage' contradicts RQ1's claim of results 'without any large-scale pretraining or dataset-specific tuning'; and (3) RQ3 reports complete-MARBLE accuracies of 87.2%/89.5%, inconsistent with Table I's 89.5%/89.8%. These concerns belong to experimental fairness and reporting accuracy, not to circularity, so the circularity score remains low.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The central claim rests on many hand-tuned constants and assumptions about dataset comparability and model training conditions. The free parameters above were all fitted to validation data or chosen by the authors, and the paper does not report sensitivity to them or release code or data.

free parameters (6)

Agent importance weights w = (3.0, 1.5, 1.2, 1.0, 1.0)
Assigned to ML, Environmental, Locational, Spatial, and Temporal agents; 'derived empirically from validation data' (Section III-E1, Eq. 14), directly weighting the vote.
Class importance factors beta = 1.2 for classes {1,4}; 1.0 for {2,3}
Boost rare classes in the weighted voting scheme (Section III-E1, Eq. 15).
SLM confidence calibration thresholds and boosts = Thresholds 0.8/0.6, boosts +0.1/+0.05, caps 0.98/0.90
Calibrate_a heuristic boosts high-confidence predictions for rare classes (Section III-C2, Eq. 10).
ML override thresholds tau_ML_high and tau_ML_corrob = 0.75 and 0.80
Set as predefined thresholds for the ML override logic (Section III-E1).
Final decision thresholds tau_coord and tie-break weight w1 = tau_coord: 0.4 rare, 0.5 common; w1: 0.7 rare, 0.5 common
Used in the final decision selection logic (Section III-F, Eq. 24-25).
Agreement boost Delta_C and confidence cap = 0.1/0.05 boost, cap 0.95
Agreement-based confidence augmentation (Section III-E1, Eq. 19 and 21).

assumptions (4)

domain assumption The severity label y can be predicted from the provided feature vector x with 4 classes for both UK and US datasets.
The paper defines Y={1,2,3,4} but UK STATS19 has 3 severity levels; no explanation of label harmonization is given (Section II, Appendix).
domain assumption The ML baseline and MARBLE's ML agent are trained under comparable data conditions.
The paper reports baselines trained on only 500 samples per class (Section V, RQ1) but does not specify the training set size for MARBLE's ML agent, making the comparison potentially unfair.
domain assumption SLM agents' free-text outputs can be reliably parsed into (class, confidence, reasoning).
Extraction function E_a uses regex or JSON parsing; the paper itself notes potential deviations in agent outputs (Section VII Limitations).
domain assumption The Smollm2-1.7B model's fine-tuning for agent-specific usage does not leak test information.
The paper states 'without any dataset-specific tuning' in RQ1, yet Table II's note says Smollm2 'was fine-tuned for agent-specific usage'; no details on the fine-tuning data are provided.

how reviews work

0 comments

Cite this review

Pith. "Pith review of MARBLE: A Multi-Agent Rule-Based LLM Reasoning Engine for Accident Severity Prediction." pith.science (2026). https://pith.science/paper/S3RH3ESH

@misc{pith2026250704893,
  author       = {Pith},
  title        = {Pith review of: MARBLE: A Multi-Agent Rule-Based LLM Reasoning Engine for Accident Severity Prediction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/S3RH3ESH}},
  note         = {Machine review of arXiv:2507.04893}
}

read the original abstract

Accident severity prediction plays a critical role in transportation safety systems but is a persistently difficult task due to incomplete data, strong feature dependencies, and severe class imbalance in which rare but high-severity cases are underrepresented and hard to detect. Existing methods often rely on monolithic models or black box prompting, which struggle to scale in noisy, real-world settings and offer limited interpretability. To address these challenges, we propose MARBLE a multiagent rule based LLM engine that decomposes the severity prediction task across a team of specialized reasoning agents, including an interchangeable ML-backed agent. Each agent focuses on a semantic subset of features (e.g., spatial, environmental, temporal), enabling scoped reasoning and modular prompting without the risk of prompt saturation. Predictions are coordinated through either rule-based or LLM-guided consensus mechanisms that account for class rarity and confidence dynamics. The system retains structured traces of agent-level reasoning and coordination outcomes, supporting in-depth interpretability and post-hoc performance diagnostics. Across both UK and US datasets, MARBLE consistently outperforms traditional machine learning classifiers and state-of-the-art (SOTA) prompt-based reasoning methods including Chain-of-Thought (CoT), Least-to-Most (L2M), and Tree-of-Thought (ToT) achieving nearly 90% accuracy where others plateau below 48%. This performance redefines the practical ceiling for accident severity classification under real world noise and extreme class imbalance. Our results position MARBLE as a generalizable and interpretable framework for reasoning under uncertainty in safety-critical applications.

Figures

Figures reproduced from arXiv: 2507.04893 by the authors.

**Figure 1.** Overview of the MARBLE architecture with a Core-Agent coordinating domain-specific agents and consensus-based inference. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

**Figure 2.** Ablation results on UK and US datasets. Panel A shows accuracy per agent-removed setting; Panel B shows relative accuracy drop. [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

**Figure 3.** ML agent confidence distributions and MARBLE outcomes on UK (A) and US (B) datasets. Bars show correct and incorrect predictions near the [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗

Figures from the paper (3 more)

**Figure 4.** Figure 4: Distribution of failure modes for the MARBLE system, based on [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]

**Figure 5.** Figure 5: Performance of MARBLE under six simulated class imbalance scenarios in the US dataset. Each setting varies the severity class distribution (top pie [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]

**Figure 6.** Figure 6: Ablation of MARBLE under coordination variants: removing con [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

45 extracted references · 33 canonical work pages

[1]

Cost of road accident fatalities to the economy,

D. Dimitriou and T. Poufinas, “Cost of road accident fatalities to the economy,” International Advances in Economic Research, vol. 22, no. 4, pp. 433–445, 2016

work page 2016
[2]

Crash injury severity prediction considering data imbalance: A wasserstein generative adversarial network with gradient penalty approach,

Y . Li, Z. Yang, L. Xing, C. Yuan, F. Liu, D. Wu, and H. Yang, “Crash injury severity prediction considering data imbalance: A wasserstein generative adversarial network with gradient penalty approach,”Accident Analysis & Prevention , vol. 192, p. 107271, 2023

work page 2023
[3]

BConvLSTM: a deep learning-based technique for severity prediction of a traffic crash,

S. R. Vinta, R. Pothuraju, M. Vijay Kumar, and G. Sai Chaitanya Kumar, “BConvLSTM: a deep learning-based technique for severity prediction of a traffic crash,” International Journal of Crashworthiness , vol. 29, no. 6, pp. 1051–1061, 2024

work page 2024
[4]

Crash severity analysis: A data- enhanced double layer stacking model using semantic understanding,

D. Yang, T. Dong, and P. Wang, “Crash severity analysis: A data- enhanced double layer stacking model using semantic understanding,” Heliyon, vol. 10, no. 9, 2024

work page 2024
[5]

Uncertainty-aware probabilistic graph neural networks for road-level traffic accident prediction,

X. Gao, X. Jiang, D. Zhuang, H. Chen, S. Wang, S. Law, and J. Haworth, “Uncertainty-aware probabilistic graph neural networks for road-level traffic accident prediction,” Transportation Research Part C: Emerging Technologies, vol. 147, p. 103982, 2023

work page 2023
[6]

Analyzing crash severity: Human injury severity prediction method based on transformer model,

Y . Jiang, X. Qu, W. Zhang, W. Guo, J. Xu, W. Yu, and Y . Chen, “Analyzing crash severity: Human injury severity prediction method based on transformer model,” Vehicles, vol. 7, no. 1, p. 5, 2025

work page 2025
[7]

The class imbalance problem in deep learning,

K. Ghosh, C. Bellinger, R. Corizzo, P. Branco, B. Krawczyk, and N. Japkowicz, “The class imbalance problem in deep learning,” Machine Learning, vol. 113, no. 7, pp. 4845–4901, 2024

work page 2024
[8]

Deep neural networks and tabular data: A survey,

V . Borisov, T. Leemann, K. Sessler, J. Haug, M. Pawelczyk, and G. Kasneci, “Deep neural networks and tabular data: A survey,” IEEE Trans. Neural Networks Learn. Syst. , vol. 35, no. 6, pp. 7499–7519, 2024

work page 2024

Show all 45 references

[9]

Traffic accident severity prediction based on interpretable deep learning model,

Y . Pei, Y . Wen, and S. P. and, “Traffic accident severity prediction based on interpretable deep learning model,” Transportation Letters , vol. 17, no. 5, pp. 895–909, 2025. [Online]. Available: https: //doi.org/10.1080/19427867.2024.2398336

2025
[10]

Interpretable traffic accident prediction: Attention spatial–temporal multi-graph traffic stream learning approach,

C. Li, B. Zhang, Z. Wang, Y . Yang, X. Zhou, S. Pan, and X. Yu, “Interpretable traffic accident prediction: Attention spatial–temporal multi-graph traffic stream learning approach,” IEEE Transactions on Intelligent Transportation Systems, 2024

2024
[11]

RFCNN: Traffic accident severity prediction based on decision level fusion of machine and deep learning model,

M. Manzoor, M. Umer, S. Sadiq, A. Ishaq, S. Ullah, H. A. Madni, and C. Bisogni, “RFCNN: Traffic accident severity prediction based on decision level fusion of machine and deep learning model,”IEEE Access, vol. 9, pp. 128 359–128 371, 2021

2021
[12]

Explainable artificial intelligence (xai): Con- cepts, taxonomies, opportunities and challenges toward responsible ai,

A. e. a. Barredo Arrieta, “Explainable artificial intelligence (xai): Con- cepts, taxonomies, opportunities and challenges toward responsible ai,” Information Fusion, vol. 58, pp. 82–115, 2020

2020
[13]

Chain-of-thought prompting elicits reasoning in large lan- guage models,

J. e. a. Wei, “Chain-of-thought prompting elicits reasoning in large lan- guage models,” in Advances in Neural Information Processing Systems (NeurIPS), vol. 35, 2022, pp. 24 824–24 837

2022
[14]

Self-consistency improves chain of thought reasoning in language models,

X. e. a. Wang, “Self-consistency improves chain of thought reasoning in language models,” in ICLR, 2023

2023
[15]

Hierarchical rein- forcement learning: A comprehensive survey,

S. Pateria, B. Subagdja, A.-H. Tan, and C. Quek, “Hierarchical rein- forcement learning: A comprehensive survey,”ACM Computing Surveys, vol. 54, no. 5, pp. 1–35, 2021. JOURNAL OF LATEX CLASS FILES, VOL. 18, NO. 9, SEPTEMBER 2020 13

2021
[16]

Hugginggpt: Solving ai tasks with chatgpt and its friends in hugging face,

Y . e. a. Shen, “Hugginggpt: Solving ai tasks with chatgpt and its friends in hugging face,” in NeurIPS, 2023

2023
[17]

A survey on llm-based multi-agent systems: workflow, infrastructure, and challenges,

X. e. a. Li, “A survey on llm-based multi-agent systems: workflow, infrastructure, and challenges,” Vicinagearth, vol. 1, 2024

2024
[18]

Recursive decomposition of logical thoughts: Framework for superior reasoning and knowledge propagation in large language models,

K. U. Qasim, J. Zhang, T. Alsahfi, and A. U. R. Butt, “Recursive decomposition of logical thoughts: Framework for superior reasoning and knowledge propagation in large language models,” 2025. [Online]. Available: https://arxiv.org/abs/2501.02026

2025 arXiv
[19]

Teaching small language models to reason,

L. C. e. a. Magister, “Teaching small language models to reason,” in ACL, 2023

2023
[20]

Symbolic chain-of-thought distillation: Small models can also “think

L. H. e. a. Li, “Symbolic chain-of-thought distillation: Small models can also “think” step-by-step,” in ACL, 2023

2023
[21]

Distilling mathematical reasoning capabilities into small language models,

X. e. a. Zhu, “Distilling mathematical reasoning capabilities into small language models,” Neural Networks, vol. 179, p. 106594, 2024

2024
[22]

Benchmarking multi-agent deep reinforcement learning algorithms in cooperative tasks,

G. e. a. Papoudakis, “Benchmarking multi-agent deep reinforcement learning algorithms in cooperative tasks,” in NeurIPS Datasets and Benchmarks Track, 2021

2021
[23]

A survey on multi-agent reinforcement learning and its application,

Z. Ning and L. Xie, “A survey on multi-agent reinforcement learning and its application,” Journal of Automation and Intelligence , vol. 2024, no. 2, pp. 73–91, 2024

2024
[24]

Least-to-most prompting enables complex reasoning in large language models,

D. e. a. Zhou, “Least-to-most prompting enables complex reasoning in large language models,” in ICLR, 2023

2023
[25]

Large language models are zero-shot reasoners,

T. e. a. Kojima, “Large language models are zero-shot reasoners,” arXiv preprint arXiv:2205.11916, 2022

2022 arXiv
[26]

Generative agents: Interactive simulacra of human be- havior,

J. S. e. a. Park, “Generative agents: Interactive simulacra of human be- havior,” in Proc. ACM CHI Conference on Human Factors in Computing Systems, 2023

2023
[28]

Self-consistency improves chain of thought reasoning in language models,

X. Wang, J. Wei, D. Schuurmans, Q. Le, E. Chi, S. Narang, A. Chowdhery, and D. Zhou, “Self-consistency improves chain of thought reasoning in language models,” in ICLR, 2022. [Online]. Available: http://arxiv.org/abs/2203.11171

2022 arXiv
[29]

Least-to- most prompting enables complex reasoning in large language models,

D. Zhou, N. Sch ¨arli, L. Hou, J. Wei, N. Scales, X. Wang, D. Schuurmans, C. Cui, O. Bousquet, Q. Le, and E. Chi, “Least-to- most prompting enables complex reasoning in large language models,” The International Conference on Learning Representations (ICLR) ,
[30]

Graph of thoughts: Solving elaborate problems with large language models,

M. Besta, N. Blach, A. Kubicek, R. Gerstenberger, M. Podstawski, L. Gianinazzi, J. Gajda, T. Lehmann, H. Niewiadomski, P. Nyczyk, and T. Hoefler, “Graph of thoughts: Solving elaborate problems with large language models,” Proceedings of the AAAI Conference on Artificial Intell...

2024 arXiv
[31]

Tree of thoughts: Deliberate problem solving with large language models,

S. Yao, D. Yu, J. Zhao, I. Shafran, T. L. Griffiths, Y . Cao, and K. Narasimhan, “Tree of thoughts: Deliberate problem solving with large language models,” Advances in Neural Information Processing Systems , vol. 36, 2023. [Online]. Available: https://arxiv.org/abs/2305.10601v2

2023 arXiv
[32]

Chain of draft: Thinking faster by writing less,

S. Xu, W. Xie, L. Zhao, and P. He, “Chain of draft: Thinking faster by writing less,” 2025. [Online]. Available: https://arxiv.org/abs/2502.18600

2025 arXiv
[33]

A survey on llm-as-a-judge,

J. Gu, X. Jiang, Z. Shi, H. Tan, X. Zhai, C. Xu, W. Li, Y . Shen, S. Ma, H. Liu, S. Wang, K. Zhang, Y . Wang, W. Gao, L. Ni, and J. Guo, “A survey on llm-as-a-judge,” 2025. [Online]. Available: https://arxiv.org/abs/2411.15594

2025 arXiv
[34]

A literature review of machine learning algorithms for crash injury severity prediction,

K. Santos, J. Dias, and C. Amado, “A literature review of machine learning algorithms for crash injury severity prediction,” Journal of Safety Research, vol. 80, pp. 254–269, 2022

2022
[35]

Predicting multiple types of traffic accident severity with explanations: A multi-task deep learning frame- work,

Z. Yang, W. Zhang, and J. Feng, “Predicting multiple types of traffic accident severity with explanations: A multi-task deep learning frame- work,” Safety Science, vol. 146, p. 105522, 2022

2022
[36]

Comparison of traffic accident injury severity prediction models with explainable machine learning,

E. Cicek, M. Akin, F. Uysal, and R. T. T. Aytas, “Comparison of traffic accident injury severity prediction models with explainable machine learning,” Transportation Letters, vol. 15, no. 9, pp. 1043–1054, 2023

2023
[37]

RFCNN: Traffic accident severity prediction based on decision level fusion of machine and deep learning model,

M. Manzoor, M. Umer, S. Sadiq, A. Ishaq, S. Ullah, H. Madni, and C. Bisogni, “RFCNN: Traffic accident severity prediction based on decision level fusion of machine and deep learning model,”IEEE Access, vol. 9, pp. 128 359–128 371, 2021

2021
[38]

Deep neural networks and tabular data: A survey,

V . Borisov, T. Leemann, K. Seßler, J. Haug, M. Pawelczyk, and G. Kasneci, “Deep neural networks and tabular data: A survey,” IEEE Transactions on Neural Networks and Learning Systems, vol. 34, no. 11, pp. 8479–8499, 2023

2023
[39]

Tabular data: Deep learning is not all you need,

R. Shwartz-Ziv and A. Armon, “Tabular data: Deep learning is not all you need,” arXiv preprint arXiv:2106.03253 , 2021

2021 arXiv
[40]

A survey on LLM- based multi-agent systems: workflow, infrastructure, and challenges,

X. Li, S. Wang, S. Zeng, Y . Wu, Y . Yang et al. , “A survey on LLM- based multi-agent systems: workflow, infrastructure, and challenges,” Vicinagearth, vol. 1, no. 1, p. 9, 2024

2024
[41]

Pre- train, prompt, and predict: A systematic survey of prompting methods in natural language processing,

P. Liu, W. Yuan, J. Fu, Z. Jiang, H. Hayashi, and G. Neubig, “Pre- train, prompt, and predict: A systematic survey of prompting methods in natural language processing,” ACM Computing Surveys, vol. 55, no. 9, pp. 1–35, 2023

2023
[42]

Language models are few-shot learners,

T. B. Brown, B. Mann, N. Ryder, M. Subbiah, J. Kaplan, P. Dhariwal, A. Neelakantan, P. Shyam, G. Sastry, A. Askellet al., “Language models are few-shot learners,” in Advances in Neural Information Processing Systems, vol. 33, 2020, pp. 1877–1901

2020
[43]

Chain-of-thought prompting elicits reasoning in large language mod- els,

J. Wei, X. Wang, D. Schuurmans, M. Bosma, E. Chi, Q. Le, and D. Zhou, “Chain-of-thought prompting elicits reasoning in large language mod- els,” arXiv preprint arXiv:2201.11903 , 2022

2022 arXiv
[44]

A countrywide traffic accident dataset,

S. Moosavi, M. H. Samavatian, S. Parthasarathy, and R. Ramnath, “A countrywide traffic accident dataset,” 2019. [Online]. Available: https://arxiv.org/abs/1906.05409

2019 arXiv
[45]

UK Road Safety: Traffic Accidents and Vehicles,

C. Tsiaras, “UK Road Safety: Traffic Accidents and Vehicles,” 2019, accessed: 2025-06-22. [Online]. Available: https://www.kaggle.com/ datasets/tsiaras/uk-road-safety-accidents-and-vehicles

2019
[2022]

Available: http://arxiv.org/abs/2205.10625

[Online]. Available: http://arxiv.org/abs/2205.10625

arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

Cost of road accident fatalities to the economy,

D. Dimitriou and T. Poufinas, “Cost of road accident fatalities to the economy,” International Advances in Economic Research, vol. 22, no. 4, pp. 433–445, 2016

work page 2016

[2] [2]

Crash injury severity prediction considering data imbalance: A wasserstein generative adversarial network with gradient penalty approach,

Y . Li, Z. Yang, L. Xing, C. Yuan, F. Liu, D. Wu, and H. Yang, “Crash injury severity prediction considering data imbalance: A wasserstein generative adversarial network with gradient penalty approach,”Accident Analysis & Prevention , vol. 192, p. 107271, 2023

work page 2023

[3] [3]

BConvLSTM: a deep learning-based technique for severity prediction of a traffic crash,

S. R. Vinta, R. Pothuraju, M. Vijay Kumar, and G. Sai Chaitanya Kumar, “BConvLSTM: a deep learning-based technique for severity prediction of a traffic crash,” International Journal of Crashworthiness , vol. 29, no. 6, pp. 1051–1061, 2024

work page 2024

[4] [4]

Crash severity analysis: A data- enhanced double layer stacking model using semantic understanding,

D. Yang, T. Dong, and P. Wang, “Crash severity analysis: A data- enhanced double layer stacking model using semantic understanding,” Heliyon, vol. 10, no. 9, 2024

work page 2024

[5] [5]

Uncertainty-aware probabilistic graph neural networks for road-level traffic accident prediction,

X. Gao, X. Jiang, D. Zhuang, H. Chen, S. Wang, S. Law, and J. Haworth, “Uncertainty-aware probabilistic graph neural networks for road-level traffic accident prediction,” Transportation Research Part C: Emerging Technologies, vol. 147, p. 103982, 2023

work page 2023

[6] [6]

Analyzing crash severity: Human injury severity prediction method based on transformer model,

Y . Jiang, X. Qu, W. Zhang, W. Guo, J. Xu, W. Yu, and Y . Chen, “Analyzing crash severity: Human injury severity prediction method based on transformer model,” Vehicles, vol. 7, no. 1, p. 5, 2025

work page 2025

[7] [7]

The class imbalance problem in deep learning,

K. Ghosh, C. Bellinger, R. Corizzo, P. Branco, B. Krawczyk, and N. Japkowicz, “The class imbalance problem in deep learning,” Machine Learning, vol. 113, no. 7, pp. 4845–4901, 2024

work page 2024

[8] [8]

Deep neural networks and tabular data: A survey,

V . Borisov, T. Leemann, K. Sessler, J. Haug, M. Pawelczyk, and G. Kasneci, “Deep neural networks and tabular data: A survey,” IEEE Trans. Neural Networks Learn. Syst. , vol. 35, no. 6, pp. 7499–7519, 2024

work page 2024