Drift detection
Machine learning models rely on accurate input data to make predictions, so it's crucial to monitor for changes in the distribution of the input data or target variable over time. If the input data or target variable changes too much from the data on which the model was trained, the model's performance may degrade when used in production. To address this issue, it's important to detect data drift as soon as possible and take appropriate action, such as building a new model or retraining the existing one.
TIM supports univariate data drift detection, where each selected variable from a dataset is checked individually for drift. The process involves splitting the dataset into two parts: reference data and test data. The reference data is used to establish the distribution of the data, and the test data is compared against the reference data to detect any drift.
Supported drift detection algorithms: