Robust OT-Guided Generative Residual Domain Adaptation for Bike-Sharing Demand Prediction under Temporal Domain Shift

In three linesGen-ROTDA, a robust optimal transport method, adapts Citi Bike demand prediction models across years (2021-2026). It transfers residuals rather than raw demand and uses a label-preserving feature generator. Gen-ROTDA achieves lowest MAE on the 2025-2026 task and outperforms non-robust OT variants under abnormal data.

Summary generated by Claude — human-verified