Scalable Real-Time Multi-Source Data Fusion Engine Utilizing Edge-Cloud Computing for Micro-Demand Forecasting in High-Velocity E-Commerce Platforms

• Ada John

  ladoke Akintola university of technology

  Author

High-velocity e-commerce platforms face unprecedented challenges in predicting micro-demand patterns at hyperlocal granularity, where traditional centralized forecasting architectures suffer from unacceptable latency and fail to capture the heterogeneous signals generated across distributed touchpoints. This study addresses the critical gap in real-time demand intelligence by designing and validating a scalable multi-source data fusion engine that leverages edge-cloud continuum computing for micro-demand forecasting. The proposed architecture integrates heterogeneous data streams—transaction records, clickstream behavior, IoT sensor telemetry from fulfillment infrastructure, and social sentiment signals—through a three-tier perception framework employing adaptive task orchestration and neural stream optimization. Methodologically, we conducted retrospective analysis on 2.4 million order events from a high-velocity platform and prospective simulation experiments comparing the proposed engine against centralized LSTM baselines and static ensemble methods. Results demonstrate that the edge-cloud fusion architecture achieves 89.4% forecasting accuracy for 15-minute interval predictions, representing a 12.7 percentage-point improvement over centralized deep learning baselines while reducing inference latency from 340ms to 47ms. The framework maintains performance stability during demand surges, with prediction error increasing only 8.3% during promotional events compared to 31.2% degradation in centralized systems. These findings establish that distributed edge intelligence with cloud-coordinated model evolution provides a theoretically grounded and practically deployable solution for micro-demand forecasting in latency-sensitive e-commerce environments, with implications extending to broader real-time analytics domains requiring sub-second decision cycles.

• pdf

Copyright (c) 2026 Ada John (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.