Privacy-Preserving Machine Learning Models for Traffic Forecasting

Mahmoud, Mahmoud N. · 2025 · ROSA P / Center for Automated Vehicles Research with Multimodal Assured Navigation (CARMEN+) Tier-1 University Transportation Center (UTC)

archive: archived pipeline: cataloged verified

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

Summary

This report addresses critical challenges in Intelligent Transportation Systems (ITS) and autonomous vehicles (AVs), specifically focusing on data privacy, model transparency, and cybersecurity. The research is motivated by the need to balance operational effectiveness with the protection of sensitive driver location data, the "black box" nature of AI decision-making in AVs, and the vulnerability of control systems to malicious attacks. The work presents four interconnected contributions designed to establish benchmarks for trustworthy AI in transportation. The study employs a multi-faceted methodological approach. First, it introduces a privacy-preserving traffic forecasting framework that integrates Inner Product Functional Encryption (IPFE) with k-anonymity mechanisms. This cryptographic scheme allows drivers to report encrypted location data to a Traffic Management Center, which aggregates the data using functional encryption keys to reveal only minimal information necessary for prediction. This aggregated data serves as input for a hybrid deep learning architecture combining Convolutional Long Short-Term Memory (Conv-LSTM), Bidirectional LSTM (Bi-LSTM), and Squeeze-and-Excitation modules to capture spatial, short-term temporal, and long-term periodic traffic dynamics. Second, the researchers apply Concept Relevance Propagation (CRP), a bias-resistant explainable AI (XAI) technique, to provide transparent, concept-level explanations for traffic detection models. Third, they leverage CRP-generated explanations to automate dataset annotation for perception models, utilizing semi-supervised learning to reduce manual labeling efforts. Finally, the study develops an explainability-guided detection framework for trojan backdoor attacks in regression-based AV steering networks, repurposing Grad-CAM and CRP to generate attribution maps that expose the rationale behind steering decisions. The findings demonstrate that the proposed privacy-preserving framework achieves high forecasting accuracy while maintaining strong privacy guarantees against collusion attacks, effectively protecting driver location data. The application of CRP enhances trust and interpretability in autonomous systems by identifying crucial latent concepts responsible for model behavior. Furthermore, models trained on auto-annotated datasets generated via CRP achieved higher mean Average Precision (mAP) scores with lower latency compared to those trained on pre-annotated datasets, indicating superior performance and efficiency. In the security domain, the explainability-guided framework achieved high detection rates for visible trojan triggers and demonstrated strong resilience against stealthy invisible variants by identifying indicators such as saliency drift and conceptual divergence. The significance of this work lies in its comprehensive approach to creating robust, transparent, and secure AI-driven solutions for transportation. By addressing fundamental challenges in data privacy, model transparency, and system security, the research supports U.S. Department of Transportation strategic priorities. It enables proactive congestion management while protecting citizen privacy, reduces barriers to perception model development through automated annotation, and strengthens cybersecurity for safety-critical applications. These contributions demonstrate practical applicability for real-world deployment, offering a foundation for trustworthy, privacy-preserving traffic management systems and enhancing regulatory compliance and public trust in autonomous vehicle technologies.

Key finding

The proposed privacy-preserving traffic forecasting framework successfully protects driver location data using functional encryption and k-anonymity while maintaining high accuracy in traffic flow predictions through a hybrid deep learning architecture.

Methodology

modeling

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 bulk_ingest_rosap on 2026-05-23 (6 acquisition events logged).

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.

• telematics crash prediction

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).

• Theoretical Contribution: computational model