Timeseries Foundation Models for Mobility: A Benchmark Comparison with Traditional and Deep Learning Models
This provides a benchmark comparison for mobility researchers, though it appears incremental as it applies existing foundation models to a specific domain.
This study evaluated TimeGPT foundation model against traditional and deep learning approaches for city-wide mobility forecasting using bike-sharing data from New York City and Vienna, finding that foundation models show potential but with identified limitations in their experimental setup.
Crowd and flow predictions have been extensively studied in mobility data science. Traditional forecasting methods have relied on statistical models such as ARIMA, later supplemented by deep learning approaches like ST-ResNet. More recently, foundation models for time series forecasting, such as TimeGPT, Chronos, and LagLlama, have emerged. A key advantage of these models is their ability to generate zero-shot predictions, allowing them to be applied directly to new tasks without retraining. This study evaluates the performance of TimeGPT compared to traditional approaches for predicting city-wide mobility timeseries using two bike-sharing datasets from New York City and Vienna, Austria. Model performance is assessed across short (1-hour), medium (12-hour), and long-term (24-hour) forecasting horizons. The results highlight the potential of foundation models for mobility forecasting while also identifying limitations of our experiments.