Abstract

Traffic congestion during peak hours and large public events is a persistent challenge in urban areas, affecting mobility, economic productivity, and quality of life. While many cities are moving towards smart, data-driven traffic management, the practical effectiveness of predictive models for event-driven traffic control remains uncertain. This thesis presents an offline, data-driven feasibility study that investigates whether ma- chine learning and time-series models can predict traffic volume patterns under different conditions, including weather and the presence of events. Using a historical traffic dataset with derived trend variables, the study applies exploratory data analysis (EDA) and two predictive approaches: ARIMA for univariate time-series forecasting and XGBoost for supervised learning based on lagged traffic volumes and contextual variables. Model performance is evaluated using standard regression and classification metrics, including Mean Absolute Error (MAE), Root Mean Square Error (RMSE), accuracy, and confusion matrices. The results show that, in this dataset, the XGBoost classifier performs only marginally better than the no-information rate, and the ARIMA model has limited forecasting accuracy during event-driven peaks. In other words, neither model predicts traffic conditions during events with sufficient accuracy for operational use. Rather than demonstrating a ready-to-deploy solution, the thesis therefore provides a critical assessment of the limitations of using simple historical data and basic model configurations for event-driven traffic management. The findings highlight the need for richer, real-time data (e.g., high-resolution event information, public transport feeds, pedestrian flows) and more advanced modeling strategies before reliable event-driven traffic control can be implemented. The study contributes by: (i) systematically documenting the gap between theoretical expectations and empirical performance of common models on a realistic dataset, and (ii) outlining concrete data and modeling requirements for future work on real-time, event-driven traffic management in smart cities.

Library of Congress Subject Headings

Traffic congestion--Management--Data processing; Predictive analytics; Machine learning

Publication Date

12-2025

Document Type

Thesis

Student Type

Graduate

Degree Name

Professional Studies (MS)

Advisor

Sanjay Modak

Advisor/Committee Member

Khalid Ezzeldeen

Campus

RIT Dubai

Plan Codes

PROFST-MS

Download

Share

COinS