Optimization of Last-Mmile Delivery Under Real-Time Traffic Variability: An Empirical Study

Thesis Abstract

Urban last-mile logistics increasingly faces volatility from real-time traffic conditions, contributing to delayed deliveries, elevated operational costs, and degraded customer satisfaction. This study addresses the problem of optimizing last-mile delivery performance under dynamic traffic variability by integrating empirical traffic data with routing and scheduling decisions. The aim is to develop and validate a framework that improves delivery reliability and efficiency despite real-time congestion, through adaptive routing, dynamic ETA prediction, and congestion-aware capacity planning. Specific objectives are (i) to quantify the impact of real-time traffic fluctuations on delivery time, fuel consumption, and vehicle utilization; (ii) to develop a Traffic-Responsive Routing Model (TRRM) that updates routes in response to live traffic signals and incident information; (iii) to construct a robust ETA forecasting module using time-series and machine learning techniques; (iv) to assess the performance of the TRRM against baseline static routing under varying congestion scenarios; and (v) to formulate practical guidelines for fleet operators on contingency planning and performance measurement. The methodology adopts a mixed-methods empirical design. The population comprises urban last-mile delivery operations in a mid-sized metropolitan area with a diversified parcel mix. A sample of 20 commercial courier fleets, totaling approximately 400 active delivery vehicles and 6,500 daily deliveries, is selected through stratified sampling to capture differences in fleet size, vehicle type, and service levels. Data collection combines (i) structured telemetry from on-board devices capturing GPS tracks, speed, acceleration, dwell times, and fuel consumption at 1-minute intervals over six months; (ii) real-time traffic data sourced from municipal road sensors, a commercial traffic API, and incident reports; (iii) operation logs including order timestamps, queue times at customers, and driver notes. An instrument suite includes a standardized driver survey for qualitative insights into routing decisions and perceived reliability; data quality checks ensure temporal alignment and sensor calibration. Analytical methods encompass quantitative and qualitative components. Time-series regression and generalized additive models quantify the relationship between traffic volatility indices and delivery performance metrics (on-time delivery rate, total travel time, fuel consumption). A Traffic-Responsive Routing Model (TRRM) is implemented as a mixed-integer linear program embedded in a rolling-horizon framework, with real-time traffic forecasts as dynamic inputs; the model’s performance is evaluated against a baseline shortest-path routing using scenario analysis across light, moderate, and severe congestion levels. ETA accuracy is assessed via MAE and RMSE, with cross-validation on hold-out periods. Sensitivity analyses examine the robustness of routing decisions to forecast errors and data latency. Additional qualitative analysis uses thematic analysis of driver logs and interviews to identify operational barriers to real-time adaptation and to validate model assumptions. The study integrates the Theory of Constraints and the Dynamic Resource Scheduling theory to anchor the design of adaptive routing and resilient delivery operations. Key expected findings include (i) measurable reductions in average delivery time and travel distance when TRRM substitutes static routing under high traffic variability, (ii) improved ETA accuracy by 15–25% relative to baseline, (iii) notable decreases in fuel consumption and vehicle wear due to congestion-aware routing, and (iv) identification of critical data latency thresholds beyond which model performance deteriorates. The study anticipates context-specific gains but also reveals limitations related to data quality, driver compliance, and information-sharing constraints among shippers and third-party carriers. The contribution to knowledge lies in (a) the empirical validation of a traffic-responsive routing framework tailored for last-mile operations, (b) integration of real-time traffic analytics with routing and ETA forecasting in an operational setting, and (c) practical guidelines for implementing adaptive logistics under real-time variability, including data governance and performance metrics. The main conclusion is that real-time traffic-aware routing and ETA forecasting substantially improve reliability and efficiency of last-mile delivery in congested urban environments when backed by high-quality data, timely forecasts, and organizational alignment. Recommendations include scaling the TRRM across networks with standardized data protocols, investing in telemetry and data-sharing practices to reduce information latency, and establishing KPI dashboards focused on on-time performance, ETA accuracy, and total cost to serve under varying traffic regimes. Suggestions for future research address incorporating stochastic driver behavior, multi-depot coordination, and city-level traffic management partnerships to further enhance resilience to traffic variability.