Extracting Sections of Simulated Driving Routes that Elicit Driving Responses Predictive of ADHD via Recursively Constructed Ensembles

Grethlein, David · 2022 · Crossref

DOI: 10.32473/flairs.v35i.130539

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of non-invasively identifying Attention Deficit Hyperactivity Disorder (ADHD) status in adolescent drivers using data from driving simulators. While motor vehicle collisions are a leading cause of death for adolescents, particularly those with cognitive impairments like ADHD, existing detection methods often rely on invasive technologies like fMRI. The authors aim to determine which specific on-road scenarios (events) elicit driving behaviors that are predictive of ADHD status, distinguishing between untreated drivers, treated drivers, and controls. This approach seeks to improve classification accuracy by filtering out non-predictive data and provide feedback to simulator designers on which scripted scenarios effectively reveal driver attention deficits. To achieve this, the authors introduce Iterative Section Reduction (ISR), an algorithm that automatically identifies spatial sections of simulated driving routes containing time series data predictive of class labels. The study utilized data from 30 participants (15 with confirmed ADHD, 15 controls) who drove four simulated routes. The dataset consisted of 91 synchronized time series channels, reduced to seven key metrics (e.g., throttle, brake, steering, velocity) down-sampled to 10 Hz. ISR recursively divides the route into overlapping sections, training classifiers on each segment to isolate those with high sensitivity to ADHD status. The authors tested two types of classifiers: similarity-based models using Dynamic Time Warping with k-Nearest Neighbors or logistic regression, and deep learning modules using a custom LSTM architecture (DeepLSTM). Hyperparameters were tuned via grid search to maximize classification accuracy and alignment with known scripted events. The results demonstrate that ISR ensembles outperformed previous global classification efforts, achieving accuracies above the prior baseline on three of the four routes. The best similarity-based ensemble achieved 62% accuracy on Drive 3, while DeepLSTM ensembles reached 50% accuracy on the same route. Crucially, the algorithm identified that events involving stop sign intersections, curves, and collision avoidance were most predictive of ADHD status. There was significant alignment between the sections identified by ISR and the scripted on-road events, particularly for curves and turns. The control group was easily distinguished, but untreated and treated ADHD drivers were often confused, suggesting medication partially mitigates risk-generating behaviors. Similarity-based models generally performed better on this small dataset, whereas DeepLSTM showed potential for scaling to larger datasets. The significance of this work lies in its ability to pinpoint specific driving scenarios that reveal ADHD-related impairments, such as jerky braking or over-steering. This provides actionable insights for designing more effective driving simulator assessments and interventions. By isolating predictive sections, the method improves classification accuracy and reduces computational load. The authors conclude that while the current findings are based on a small dataset, the ISR algorithm successfully isolates discriminatory spatial regions, offering a robust framework for future research into non-invasive ADHD screening and driver safety assessment.

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.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-09; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• drowsiness detection algorithms
• distraction detection algorithms

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).

• Methodological Resource: tool software, validation psychometrics
• Theoretical Contribution: computational model