
Process of removing noise or artifacts from data to reveal true signals.Data cleaning is the process of detecting and correcting (or removing) corrupt, inaccurate, or irrelevant parts of a dataset. In physics, raw measurements are never perfect: detectors have noise, environmental interference creeps in, and systematic errors distort signals. Cleaning separates the true physical signal from these artifacts, making subsequent analysis reliable. It is the unglamorous but essential first step in almost every experimental and observational study, from quantum imaging to cosmology.
The term became prominent with the rise of digital data processing in the mid-20th century, but the practice is as old as experimental science. Early astronomers manually removed bad observations; modern automated pipelines emerged with computers in the 1960s–70s.
Typical methods include filtering (e.g., removing frequency bands dominated by noise), outlier rejection (discarding points that deviate wildly from expected behavior), and calibration (using known references to correct for instrument response). More advanced approaches use statistical models or machine learning to identify and subtract artifacts, such as neural networks that distinguish real events from detector glitches. The key is to remove noise without removing real signal, which requires understanding the noise's characteristics and the physics of the measurement.
Used in every experimental physics lab, from LIGO's gravitational-wave detectors (cleaning seismic noise) to quantum optics (removing dark counts) and astrophysics (separating foregrounds from cosmic microwave background).
In the LIGO gravitational-wave detectors, data cleaning includes vetoing times when airplanes fly overhead, because their radar signals can mimic gravitational-wave events.






