What role does data quality play in machine learning?

Example Answer

The way I'd explain it in an interview is this: Data quality sets the ceiling on model quality. If the labels are wrong, the features are stale, or the training data does not represent real production behavior, the model will learn the wrong thing no matter how sophisticated the architecture is.

I usually break data quality into a few concrete dimensions: correctness, completeness, consistency, timeliness, and representativeness. That lets me reason about real failure modes like duplicate events, train-serve skew, missing values that are not random, and label definitions that changed quietly over time. Those issues often hurt performance more than model choice does.

In practice, strong teams treat data quality as an ongoing system, not a preprocessing task. They validate inputs, monitor drift, audit labels, and make data contracts explicit so model performance is not resting on hidden assumptions.

Common Poor Answer

A weak answer says data quality is important but never explains which failure modes matter or how bad data actually shows up in model behavior.