Model using Sampling Strategy - Down/Up etc and its Real Time Deployment

sampling

#1

Using one of the sampling strategy, say down/up/synthetic sampling, by which we better the event rate in the imbalanced data and hence improve on accuracy, mostly True Positives. Of-course, this strategy helps in good prediction on train and test data.

Ask: In this scenario, when this model is deployed into production where event rate is still pretty low, how good your predictions are in real time?