Abstract
Airline demand, a key component of the aviation industry, is difficult to predict due to demand stochasticity and affects other problem areas of airline operations. Because of this stochasticity, various statistical and artificial intelligence models have been employed to predict airline demand. However, these attempts to forecast demand have often proved inaccurate, leading to difficulties in fleet planning and other operational problems that airlines must solve subsequently. Recognizing the insufficiency of previous models in the airline industry, this project provides an AI-based solution using a Temporal Fusion Transformer (TFT) designed for time series forecasting. This integrated approach addresses cyclical patterns, seasonal variations, and unexpected disruptions in a unified mathematical framework. We validate our approach using historical domestic route data from the US Bureau of Transportation Statistics. The model outperforms traditional methods, we present that using Mean Absolute Percentage Error (MAPE) and more realistic confidence intervals that properly quantify prediction uncertainty. The focal point of this research is the improvement of performance metrics via stochastic integration.
Recommended Citation
Maru, Vatsal; Shekhar, Gaurav; and Ponzo, Kade, "Aircraft Route Demand Forecasting using Temporal Fusion Transformers (TFT)" (2025). MWAIS 2025 Proceedings. 6.
https://aisel.aisnet.org/mwais2025/6