Short-Term Forecasting and Anomaly Detection of Melbourne CBD Pedestrian Flows Using Hourly Sensor Counts

School of Computing and Information Systems, Faculty of Science, The University of Melbourne, Melbourne, Australia

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

This research aims to analyze the open-source pedestrian counts from the city of Melbourne to identify temporal patterns, forecastability, and anomalous pedestrian activity within the Melbourne central business district (CBD). This research utilizes hourly pedestrian counts from 23rd January 2024 to 22nd January 2026, which then form the basis for city-level time series indicators. Three main analytical approaches will be adopted for the purpose of this research: a weekly diurnal heatmap, a series representing the total pedestrians per day, and an hourly anomalous detection using a z-score method. A simple model will also be adopted to identify forecastability. The results clearly indicate the presence of temporal patterns within the data set, including strong intraday rhythms and notable differences between weekdays and weekends. The simple model was effective in capturing the main temporal patterns observed. It had good overall forecastability for the test set. The results from the anomaly detection highlighted a small number of extreme deviations in the data. These deviations were mainly positive short-term spikes. It is evident that the results demonstrate that time series analysis is a viable tool for understanding normal and abnormal pedestrian activity in the Melbourne CBD.