Information leakage

Information leakage is one of the common and important errors in data handling during all machine learning applications, including those in radiology. Briefly, it means the incomplete separation of the training, validation, and testing datasets, which can significantly change the apparent performance of the algorithmic method. 

Since data overlap between datasets is a critical biasing factor, it is crucial to split data at the beginning of the study, before proceeding with any further steps (e.g. feature extraction) as these can result in data leakage 1.
 

Artificial intelligence

Article information

rID: 83268
Synonyms or Alternate Spellings:

ADVERTISEMENT: Supporters see fewers/no ads

Updating… Please wait.

 Unable to process the form. Check for errors and try again.

 Thank you for updating your details.