AI data could be tainted even as it’s being cleaned

Risk USA: Expert says even touching raw data could lead to loss of context

Data cleansing efforts should be properly documented, says Capital One's Hanif

Companies cleaning the data they’re using for their machine learning models could unintentionally adulterate it in the process, one expert has said.

“Anytime you touch the data before it enters your algorithm, there is absolutely always the risk that it removes something that has contextual information, and you don’t know it yet,” said Zachary Hanif, principal machine learning engineer at Capital One, who spoke on a panel on data science at the Risk USA conference in New York on November 9.

Th