What does “noise” mean when training models with real-world data?
2 Likes
Noise refers to irrelevant or inaccurate data that masks real patterns - like mislabeled images, missing values, or outlier behavior. Too much noise reduces model accuracy and causes unstable predictions.
1 Like
It’s randomness or inconsistency the model can’t genuinely learn from. Filtering noise with preprocessing, normalization, or better labeling improves performance and generalization.