Predicting Vehicle Type Using NGSIM US-101
Year: 2026
Type: software
Abstract: Supervised machine-learning models (Logistic Regression, Decision Tree and Random Forest) classify motorcycles, passenger cars and trucks from the NGSIM US-101 traffic trajectory dataset. Method and evaluation. The notebook aggregates 10 Hz trajectories into 6,101 vehicle-level observations to reduce leakage from repeated observations of the same vehicle. It compares the three model families with attention to class imbalance and macro-F1, and includes a dimensions-only versus full-feature ablation study. Illustrated engineering documentation. The companion case study explains the data preparation, modelling workflow, engineering rationale and methodological limitations. It brings together 14 figures extracted from the public notebook and two newly authored explanatory diagrams. These diagrams describe the analytical workflow; they are not additional experimental results. Illustrated case study: explanation, diagrams and source figures Source repository and reproducible notebook Detailed engineering guide Scope and provenance. This record archives software version v1.0.0. The companion documentation may evolve independently of the archived release. The work is an analysis of the NGSIM US-101 dataset; results should be interpreted in that dataset context rather than as evidence of performance in a deployed traffic-enforcement system.
Author: Mahyoub, Mohammed
Language: English
Subfield: Transportation
Field: Social Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7214711212