Professional Data EngineerOperationalizing machine learning modelsEasy
A retail company uses a machine learning model to optimize inventory levels. The model was trained on historical sales data. Over time, new product lines are introduced, and consumer purchasing habits shift due to market trends. The current model's performance has started to degrade, leading to suboptimal inventory recommendations. What is the most likely cause of this degradation?
- ALabel drift
- BModel bias
- CData leakage
- DFeature drift
Show answer & explanationAnswer & explanation
Correct answer: D. Feature drift
Feature drift occurs when the statistical properties of the input features to the model change over time, which directly impacts the model's ability to make accurate predictions based on its original training.
Why the other options are wrong
- A. Label drift refers to changes in the relationship between features and labels, or changes in the labels themselves, not directly the input features.
- B. Model bias refers to systematic errors in the model's predictions, often due to unrepresentative training data, but it doesn't typically manifest as a degradation over time due to changing market conditions.
- C. Data leakage is when information from outside the training dataset is used to create the model, leading to overly optimistic performance during training, not a degradation over time in production.
Feature Drift
A phenomenon where the statistical properties of the input features to a machine learning model change over time, leading to degraded model performance.
- Also known as covariate shift.
- Can be caused by real-world changes (e.g., market trends, new data sources).
- Requires model retraining or adaptation.
Memory trick: Drifting features make models lose their way.