Crash-severity assessment on NEISS Whole
80.4AccuracyOlogit
Evaluation results
| Method | Links | ||
|---|---|---|---|
| OlogitBackbone LLM=N/A, Number of samples=10592026.01 | 80.4 | 22.3 | |
| MLP + feature selectionBackbone LLM=N/A, Number of samples=10592026.01 | 79.1 | 27.7 | |
| MLPBackbone LLM=N/A, Number of samples=10592026.01 | 78.7 | 31.2 | |
| TransportAgentBackbone LLM=GPT4o-mini, Number of samples=10592026.01 | 74.84 | 53.4 | |
| AutoGenBackbone LLM=GPT4o-mini, Number of samples=10592026.01 | 62.04 | 31.3 | |
| 1 k-shot (vanilla LLM)Backbone LLM=GPT4o-mini, Number of samples=10592026.01 | 46.7 | 25.06 | |
| CoTBackbone LLM=GPT4o-mini, Number of samples=10592026.01 | 41.4 | 27.97 | |
| OlogitMethod Category=Traditional Econometric Model2026.01 | 0.804 | 0.223 | |
| MLP + feature selectionMethod Category=Traditional ML Models, Feature Selection=true2026.01 | 0.791 | 0.277 | |
| MLPMethod Category=Traditional ML Models2026.01 | 0.787 | 0.312 | |
| TransportAgentBackbone=GPT-3.5-turbo, Method Category=Hybrid LLM-ML Agent Framework2026.01 | 0.7641 | 0.449 | |
| AutoGenBackbone=GPT-3.5-turbo, Method Category=Augmented LLM-based Reasoning2026.01 | 0.6185 | 0.259 | |
| 1 k-shot (vanilla LLM)Backbone=GPT-3.5-turbo, Reasoning Strategy=1 k-shot, Method Category=Pure LLM-based Reasoning2026.01 | 0.467 | 0.1679 | |
| CoTBackbone=GPT-3.5-turbo, Reasoning Strategy=Chain-of-Thought, Method Category=Pure LLM-based Reasoning2026.01 | 0.4198 | 0.3605 |