PREDICTIVE MODELING FOR ONLINE FOOD DELIVERY SYSTEMS: DEMAND FORECASTING AND ETA ESTIMATION UNDER CONTEXTUAL UNCERTAINTY

Abstract

Online food delivery services have grown significantly in recent years. This growth is primarily driven by shifts in consumer behavior and the increasing adoption of digital platforms. As demand grows, optimizing operational processes becomes critical, particularly under uncertain conditions. Online food delivery systems operate as inherently stochastic end-to-end cycles. They begin with customer order placement and involve platform-level coordination, restaurant preparation, courier assignment, and last-mile delivery. Each stage is influenced by dynamic and uncertain factors such as demand fluctuations, traffic conditions, weather variability, and operational congestion. This thesis develops a predictive modeling framework for improving operational intelligence in online food delivery systems. Dual forecasting objectives are considered: short-term demand prediction and delivery-time (ETA) estimation. For demand forecasting, the study evaluates advanced machine learning models across multiple temporal resolutions. Results demonstrate that predictive performance varies by aggregation level, with short-term forecasting providing the greatest operational value for real-time dispatching and staffing decisions. The findings also highlight that model effectiveness depends on alignment between temporal structure and algorithmic architecture. For ETA estimation, the research adopts a structured modeling pipeline beginning with a diagnostic internal-data baseline. The results reveal that internal operational features alone have a limit. By reconstructing geospatial information and integrating contextual variables such as real-time traffic and weather data, the proposed framework achieves significant improvements in accuracy and reliability. A key insight is that contextual data enrichment can achieve better predictive gains than increasing model complexity. Overall, the thesis contributes a unified and scalable framework for predictive analytics in online food delivery systems under contextual uncertainty. By integrating demand forecasting and ETA modeling, this work advances data-driven decision support for dynamic digital logistics platforms.

Date of Award	2026
Original language	American English
Awarding Institution	HBKU College of Science and Engineering