Title:Back To The Future: A Hybrid Transformer-XGBoost Model for Action-oriented Future-proofing Nowcasting

Abstract:Inspired by the iconic movie Back to the Future, this paper explores an innovative adaptive nowcasting approach that reimagines the relationship between present actions and future outcomes. In the movie, characters travel through time to manipulate past events, aiming to create a better future. Analogously, our framework employs predictive insights about the future to inform and adjust present conditions. This dual-stage model integrates the forecasting power of Transformers (future visionary) with the interpretability and efficiency of XGBoost (decision maker), enabling a seamless loop of future prediction and present adaptation. Through experimentation with meteorological datasets, we demonstrate the framework's advantage in achieving more accurate forecasting while guiding actionable interventions for real-time applications.