VFlash: Fast, Accurate Visual Planning via Flash Primitive Decoding
u_Unrolled
Easy10 XPmetricsimbalance

Why is 99% accuracy meaningless on imbalanced data?

1% of transactions are fraud. A model that always predicts "not fraud" has what accuracy, in percent?

%
Stuck? Show the solution

Answer: 99 %

It is right on the 99% of legitimate transactions and catches no fraud at all. On imbalanced data, look at precision, recall and PR-AUC instead.

Learn the idea behind itML Evaluation — Measuring Models Without Fooling Yourself