medium10 pts · first answer countsmonitoringdrift

Labels arrive weeks later

A loan default model only gets true labels months after prediction. How can you detect problems sooner?

A.You cannot monitor anything until labels arriveB.Monitor proxies: input feature drift, prediction distribution shift, missing values and early indicators such as first missed paymentC.Retrain every day on unlabelled dataD.Compare predictions with a random number generator

← Safe retries in pipelines Computing PSI for two bins →