In this installment of the AI Simplified video series Jake Shaver, Program Manager at DataRobot, talks about target leakage, one of the most challenging issues when developing a machine learning model.
“Target leakage is a problem because if you use a mode that contains leakage, it may look good during the training phase, but it will make wildly inaccurate predictions once you finally get to using it in a production environment.” — Jake Shaver
Target leakage is essentially when you’re using data from the future in order to try and predict the past. Unsure of how this can happen? You’re not alone. Watch the AI Simplified: Target Leakage video below for a high level introduction to the topic along with a useful analogy and helpful use cases:
Want more AI Simplified? Visit our YouTube channel for more videos!