An AI-Enhanced Edge-to-Lakehouse Architecture for Real-Time Safety Analytics in Last-Mile Delivery Fleets
DOI:
https://doi.org/10.63282/3050-9246/ICRTCSIT-133Keywords:
Last-Mile Delivery, Fleet Safety, Edge Computing, Lakehouse Architecture, Data Pipelines, Big Data, Ai Analytics, Driver MonitoringAbstract
The rapid expansion of safety-critical signals from last-mile delivery fleets introduces possibilities and constraints to existing real-time risk mitigation strategies. The legacy data lake approach, committed to batch processing, and disjointed IoT telematics frameworks, often cannot deliver low-latency, audit-trail-ready, insights for driver distraction detection, accident avoidance, and regulatory compliance or investigation. Our paper develops an integrated Edge-to-Lakehouse design that combines in-vehicle preprocessing, streaming ingestion, ACID-compliant storage, and integration with a feature store. We present issues in current practice that limit addressing safety risk including limited uptake of AI-enabled safety pipelines, lack of real-time analytics, heterogeneity, schema drift, lack of auditability, and limited observability and propose a longitudinal system design that seeks to remedy these issues. The methodology deliberately partitions workloads to limit cloud resource usage while balancing latency and cost and include guidance for handling evolved schemas and longitudinal lineage metadata. Operational observability is incorporated as a design principle. Evaluation metrics will include end-to-end latency, feature freshness, predictive / prescriptive model performance (e.g. AUC / F1), and pipeline reliability across all workflows. We propose a system for predictive and prescriptive safety analytics that move fleets beyond descriptive dashboard technologies to do more proactive safety accident management
Downloads
References
[1] P. Visconti, G. Rausa, C. Del-Valle-Soto, R. Velázquez, D. Cafagna, and R. De Fazio, “Innovative Driver Monitoring Systems and On-Board-Vehicle Devices in a Smart-Road Scenario Based on the Internet of Vehicle Paradigm: A Literature and Commercial Solutions Overview,” Sensors, vol. 25, no. 2, p. 562, Jan. 2025. https://doi.org/10.3390/s25020562
[2] W. Ding, C. Xu, M. Arief, H. Lin, B. Li, and D. Zhao, “A Survey on Safety-Critical Driving Scenario Generation- A Methodological Perspective,” IEEE Transactions on Intelligent Transportation Systems., vol. 24, no. 6, pp. 6789–6806, Jun. 2023. https://doi.org/10.48550/arXiv.2202.02215
[3] X. Zhou, R. Ke, H. Yang, and C. Liu, “When Intelligent Transportation Systems Sensing Meets Edge Computing: Vision and Challenges,” IEEE Canadian Journal of Electrical and Computer Engineering, vol. 48, no. 4, pp. 123–135, Dec. 2023. https://doi.org/10.3390/app11209680
[4] Selvaraj, P. Sivathapandi, and D. Venkatachalam, “Artificial Intelligence-Enhanced Telematics Systems for Real-Time Driver Behaviour Analysis and Accident Prevention in Modern Vehicles,” J. Artif. Intell. Res., vol. 3, no. 1, pp. 198–239, Jun. 2023. https://thesciencebrigade.com/JAIR/article/view/368
[5] W. Liang, A. Taofeek, A. Rajuroy, and M. Blessing, “Developing Scalable and Secure Feature Stores in Real-Time Machine Learning Pipelines,” J. Artif. Intell. Res., vol. 4, no. 2, pp. 45–78, Jul. 2025. https://www.researchgate.net/publication/393622612
[6] A. Awad, J. Traub, and S. Sakr, “Adaptive Watermarks: A Concept Drift-based Approach for Predicting Event-Time Progress in Data Streams,” in Proc. 22nd Int. Conf. Extending Database Technology (EDBT), Lisbon, Portugal, Mar. 2019, pp. 622–625. https://www.researchgate.net/publication/332570283
[7] N. Janssen, T. Ilayperuma, J. Jayasinghe, F. Bukhsh, and M. Daneva, “The evolution of data storage architectures: examining the secure value of the Data Lakehouse,” J. Data Inf. Manag., vol. 6, no. 4, pp. 309–334, Aug. 2024. https://doi.org/10.1007/s42488-024-00132-1
[8] M. R. Pulicharla, “Detecting and addressing model drift: Automated monitoring and real-time retraining in ML pipelines,” World J. Adv. Res. Rev., vol. 3, no. 2, pp. 147–152, Sep. 2019. [Online]. Available: https://doi.org/10.30574/wjarr.2019.3.2.0189
[9] C. Bogart, R. Chhajer, B. Singh, T. Fontana, and M. Sakr, “PlantD: Performance, Latency Analysis, and Testing for Data Pipelines — An Open Source Measurement, Testing, and Simulation Framework,” presented at Proc. CODS COMAD 2025, Lisbon, Portugal, Jan. 2025. https://doi.org/10.48550/arXiv.2504.10692
[10] A. A. Khan, O. Chaudhari, and R. Chandra, “A review of ensemble learning and data augmentation models for class imbalanced problems: Combination, implementation and evaluation,” Expert Syst. Appl., vol. 222, Art. no. 122778, 2023. https://doi.org/10.1016/j.eswa.2023.122778
[11] E. Breck, S. Cai, E. Nielsen, M. Salib, and D. Sculley, “The ML Test Score: A Rubric for ML Production Readiness and Technical Debt Reduction,” in Proc. IEEE Int. Conf. Big Data, Boston, MA, USA, Dec. 2017 https://research.google/pubs/the-ml-test-score-a
[12] A. Reuel, A. Hardy, C. Smith, M. Lamparth, M. Hardy, and M. J. Kochenderfer, “BetterBench: Assessing AI Benchmarks, Uncovering Issues, and Establishing Best Practices,” presented at NeurIPS 2024, Dec. 2024. https://doi.org/10.48550/arXiv.2411.12990
