Predicting Future Pedestrian Movement Patterns: An LSTM Approach Using Historical Footfall Data from Dublin

Accurate forecasting of pedestrian footfall is essential for understanding temporal patterns in urban mobility. Existing statistical and machine learning approaches, like regression-based models, struggle to capture the nonlinear temporal dependencies and complex seasonal patterns common in urban pedestrian datasets, hence creating a need for more robust sequence learning methods. To address this limitation, this study implements and evaluates a Long Short-Term Memory (LSTM) neural network to predict daily pedestrian counts across Dublin City Centre using three years of sensor data (2022–2024) collected via PYRO-BOX counters. Hourly data were aggregated to daily totals, preprocessed to handle missing values, and assessed for stationarity. A 30-day sliding window approach was applied to construct sequential input-output pairs, and the dataset was partitioned into 80 per cent training and 20 per cent testing subsets. The LSTM model was trained with dropout regularisation and early stopping to prevent overfitting. Predictive performance was evaluated using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), and R-squared (R²). The model demonstrated a strong predictive performance, with an MAE of 26,069.72, an MAPE of 8.82 per cent, and an R² of 0.7778, capturing both seasonal trends and complex temporal dependencies. A 30-day forecast starting January 1, 2025, closely aligned with historical footfall data, demonstrating the model’s ability to project future values from historical patterns. These findings contribute both theoretically and practically by demonstrating the effectiveness of deep learning based temporal modelling for urban pedestrian dynamics and by providing an evidence-based forecasting framework that can support urban planning, transport policy decisions, such as crowd management and event planning.