What term describes the gradual change in data characteristics over time that can degrade model performance?

The term that specifically describes the gradual change in data characteristics over time, which can negatively impact model performance, is "data drift." Data drift occurs when the statistical properties of the input data to a model change, leading to a scenario where the model's predictions become less accurate or relevant as the relationship between the input variables and the target variable evolves. This phenomenon is particularly important to monitor in machine learning and AI systems, as it can signal that the model needs to be updated or retrained with more relevant data to maintain its effectiveness.

In contrast, "data decay" often refers to the diminishing relevance or quality of data over time, particularly in the context of information that becomes outdated rather than changes in statistical properties. "Data loss" refers to the scenario where data becomes unavailable or corrupted, providing a different concern altogether. Lastly, "data shift" may sometimes be used interchangeably with data drift; however, it is less commonly used and might imply a more sudden change rather than a gradual one. Therefore, the most precise term for the situation described in the question is indeed data drift.