The Contradiction of Overall Accuracy Gains vs. Segment-Level Failure: What is 'Simpson's Paradox' That Could Destroy Your Model?

On the day your team introduced a new machine learning model, your manager reported that 'accuracy improved across all ...