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

The interplay between past, present, and future is a central theme in the iconic movie Back to the Future, where small alterations in past events have profound, cascading effects on the future [1]. This concept mirrors the intricate and often non-linear relationships in real-world systems, where predictions about the future are not merely passive observations but active drivers of current decisions and behaviors. In the film, the characters reshape their present and future by altering past events, reflecting the power of temporal causality--the idea that events in time are interconnected, and that actions taken now have consequences for the future. In much the same way, effective nowcasting--predicting short-term outcomes like weather, natural hazards, health events, or traffic patterns--should not only anticipate what will happen but also incorporate how those predictions can influence present decisions and conditions. Traditional nowcasting methods, however, often focus exclusively on making predictions about future states without considering the active feedback loop that can be created by those predictions [2].