Road Travel Time Saved for Emergency Vehicles: A Journey-Level Analysis of GRU Predictive Routing

Author

Freshia Njoki Macharia, Gilbert Langat, Stephen Mageto


Abstract

 A published emergency-routing framework that joins a gated recurrent unit forecast, time-dependent A*, and a remaining-time threshold reported mean travel-time reductions of 16.19% under a simulated incident, 1.03% in peak hour, and 0.01% off-peak (Macharia et al., 2026). Those means were taken across matched journeys and, in the regime tables, across four threshold settings on the same origin–destination pairs. This article keeps that framework fixed and reports, at the operational threshold δ = 0.20, how road travel time is distributed across 75 journeys in each regime. A journey counts as a win or a loss only when realised time differs from snapshot Dijkstra by more than five seconds. Under the incident regime the framework was faster on 71 of 75 journeys, slower on none, and tied on four. The median saving was 83.1 seconds and the mean was 127.7 seconds (paired d = 1.22; 95% bootstrap interval 104.7 to 153.0 seconds). Fifty-two journeys saved at least one minute, and 28 saved at least two. The time-weighted reduction was 17.6%. Short incident journeys saved 64.4 seconds on average and long ones saved 220.5 seconds (r = 0.77). Ranked by seconds saved, 17 of the 75 incident journeys account for half of the road time removed. In peak hour the median saving was zero seconds: 22 journeys were faster, 46 stayed within five seconds, and 7 were slower, one of them by 68 seconds. Off-peak removed 10.7 seconds in total across 75 journeys. Against a reactive snapshot, the framework was faster on 11 incident journeys, tied on 60, and slower on 4; the median advantage was zero seconds and the mean advantage was 4.0 seconds. Raising δ from 0.05 to 0.20 left the incident win count at 71 and moved the 90th percentile of seconds saved from 282.5 to 263.2. The published 16.19% describes the incident journeys. Peak-hour and off-peak medians are zero.


Keywords

emergency vehicle routing; journey-level evaluation; practical significance; travel-time distribution; predictive routing; heterogeneity



Full Text:

Download Paper PDF


References


  • Chowdhury, A., Kaisar, S., Khoda, M. E., Naha, R., Khoshkholghi, M. A., & Aiash, M. (2023). IoT-based emergency vehicle services in intelligent transportation system. Sensors, 23(11), 5324. https://doi.org/10.3390/s23115324
  • Feng, J., Guo, G., Wang, J., Liu, X., Liu, Z., & Ding, Y. (2024). EVHF-GCN: An emergency vehicle priority scheduling model based on heterogeneous feature fusion with graph convolutional networks. IEEE Access. https://doi.org/10.1109/ACCESS.2024.3349591
  • Jiang, W., & Luo, J. (2022). Graph neural network for traffic forecasting: A survey. Expert Systems with Applications, 207, 117921. https://doi.org/10.1016/j.eswa.2022.117921
  • Kang, K., Jones, M. T., Armstrong, K., Avery, S. N., McHugo, M., Heckers, S., & Vandekar, S. (2023). Accurate confidence and Bayesian interval estimation for non-centrality parameters and effect size indices. Psychometrika, 88(1), 253–273. https://doi.org/10.1007/s11336-022-09899-x
  • Liu, X. S. (2023). Bias correction for Cohen’s d. The Journal of General Psychology. https://doi.org/10.1080/00221309.2023.2172545
  • Macharia, F. N., Langat, G., & Mageto, S. (2026). Design and evaluation of a GRU-based predictive routing framework for emergency vehicles. International Journal of Research Publication and Reviews, 7(8), Article IJRPR69191. https://doi.org/10.55248/gengpi.07.0826.2618
  • Qi, P., Pan, C., Xu, X., Wang, J., Liang, J., & Zhou, W. (2025). A review of dynamic traffic flow prediction methods for global energy-efficient route planning. Sensors, 25(17), 5560.  https://doi.org/10.3390/s25175560

 

Share your valuable work from Social Media Buttons