🤖 The 99% accurate model that caught nothing

An intern built a fraud detection model for a payments company. It is 99% accurate on test data. In its first week live, it did not flag a single real fraud.

ஒரு intern payments நிறுவனத்துக்கு fraud detection model செய்தார். Test data-வில் 99% accuracy. ஆனால் live-ல் முதல் வாரத்தில் ஒரு fraud-ஐயும் பிடிக்கவில்லை.

Step 1 of 3

Only 1 in 100 transactions is fraud. Why can a useless model be 99% accurate?