Abstract

Accurate air travel demand forecasting is critical for airline profitability, yet predicting the market stimulation resulting from new capacity additions remains a significant challenge. This research addresses this gap by developing and evaluating a series of deep learning models to forecast incremental passenger demand following capacity expansion. Utilizing a longitudinal dataset of Canadian international Origin-Destination (O-D) pairs from 2014 to 2024, this study integrates passenger flow data with airline schedule information, including engineered features to capture capacity growth and new route introductions. Four distinct neural network architectures were implemented and compared: a baseline Artificial Neural Network (ANN), a 1D Convolutional Neural Network (CNN), a Long Short-Term Memory (LSTM) network, and a Gated Recurrent Unit (GRU) network. The GRU model achieved the lowest Mean Absolute Percentage Error (MAPE) at 5.23%, outperforming all other models. In contrast, the ANN and CNN models produced higher error rates, indicating limited ability to capture temporal dependencies. The GRU maintained consistent performance even during the volatile 2020–2024 period, suggesting its suitability for forecasting under disrupted market conditions. These results support the use of recurrent architectures for modeling demand stimulation and provide a data-driven framework for route planning, revenue optimization, and resource allocation in the airline industry.