Concept Drift

Concept drift is a specific type of model drift where the fundamental relationship between input features and the target variable changes. Unlike simple data drift, where the input values change, concept drift means the logic governing the market has shifted.

In cryptocurrency, this might occur if a major protocol upgrade fundamentally changes how a token is valued. The model's internal logic, which worked previously, is now based on a false premise.

Identifying concept drift requires deep analysis of model performance metrics and a strong understanding of the underlying asset. It necessitates a complete reassessment of the model's features and potentially a redesign of the strategy.

It is one of the most difficult challenges to solve in automated financial modeling.