Abstract: Traffic flow prediction is a crucial component of developing intelligent transportation systems in smart cities. The core goal is to investigate the data engineering approaches and methodologies that support traffic flow prediction. Structured but diverse traffic flow data sets generated by sensor networks and other sources are ingested, processed, and stored in data platforms that enable operational real-time or near-real-time predictions and alternative exploratory offline analyses. Effective data engineering process solutions for traffic flow prediction require the careful design of all stages—from data sources to model application—because the dynamic and still little-understood nature of traffic flow data makes dedicated traffic prediction models sensitive to a variety of factors, including data quality, time frame, spatial resolution, external data, feature representation, and algorithm selection. Consideration of these factors can yield appropriate solutions across data scenario alternatives: data from a limited number of senors, real-time prediction with external data to better represent special events, online learning for concept drift adaptation, and feature engineering with new perspectives or resources.

Traffic flow prediction represents an operational application in the complex and multidisciplinary scenarios of a smart city. Intelligent transport systems rely on timely and accurate real-time predictions to optimize vehicle distribution, reduce waiting times, increase passenger satisfaction, and enable vehicle tracking. Such predictions also support external decision-making processes that require support from an intelligent system or subsystem. Usually classified as time-series forecasting, traffic flow prediction aims to inferring future values of a time-ordered series generated by one or more object through a range sensor, such as microwave, loop, infrared, or video camera sensor. The tasks for traffic flow prediction comprises sensor networks and the incoming traffic flow data streams, as well as multiple connected external data source that contribute to broaden the representation of predicted-related phenomena—e.g. weather conditions, official event schedule—and support data fusion.

Keywords: Traffic flow prediction, Smart cities, Data engineering, Spatio-temporal data, Intelligent transportation systems (ITS), Big data analytics, Internet of Things (IoT), Real-time data processing, Data pipelines, Feature engineering, Graph neural networks (GNN), Deep learning, Time series forecasting, Data fusion (multi-source integration), Edge computing.

Downloads: | DOI: 10.17148/IJIREEICE.2021.91222