oversampling

In digital signal processing, this term describes a technical strategy to shift quantization noise to higher frequencies, allowing for a simpler analog filter. It is a precise technical term used primarily by engineers and audio specialists to describe a method of increasing fidelity.

In the context of data science and machine learning, the term refers to a specific remedy for class imbalance. It is often contrasted with undersampling, where the majority class is reduced. Users should be aware that while oversampling balances a dataset, it carries a high risk of overfitting if the same minority examples are simply duplicated.