Orchestrating Urban Footfall Prediction: Leveraging AI and batch-oriented workflow for Smart City Application

Abstract

This paper explores development and deployment of a smart city prediction system, demonstrating this capability on data generated by footfall counting sensors. Presented approach integrates classical machine learning (ML) techniques with process orchestration framework Apache Airflow. The architecture is designed to handle datasets in periodic batches, ensuring updates are regularly integrated into the prediction system and new predictions are created at every increment. Our work demonstrates ease at which similar systems can be developed, given sufficient volume of data and availability of compute power. This approach highlights that increasing number of smart sensors, availability of proven ML techniques and modern processing frameworks create a critical mass for proliferation of real-time forecasting solutions. Our results indicate that the developed system is effective in predicting footfall patterns, a variable that can be instrumental in applications such as traffic control, resource allocation, public safety, and urban planning. Used methodology is not limited to footfall data, and can be applied to other timeseries datastreams, making it a versatile tool for smart city context. Showcasing practical implementation and benefits of the system, the paper contributes to the ongoing efforts in developing a class of digital urban infrastructure.