Target Leakage

Target leakage occurs when a model relies on data that will not be known at the prediction time. The model is then less accurate when used on actual data. It’s crucial to recognize potential sources of target leakage throughout your project. On this mission, you will study a set of scenarios that demonstrates how target leakage can be introduced throughout the model building process. You will also learn methods to prevent target leakage and rectify the issue if it occurs. By studying these scenarios, you’ll be better able to head off target leakage in your models.

Mission format and duration: self-paced, hands-on, 1.5 hours

Upon completion of this mission, you will be able to: 

• Detect potential sources of target leakage during data collection
• Prevent the introduction of target leakage during feature engineering and data partitioning
• Safeguard against target leakage during training and tuning

Who should complete this mission?

• Business Analysts
• Citizen Data Scientists
• Data Scientists

Before embarking on this mission, you should complete one of the following:

• Starter Quest — appropriate to your role (self-paced mission)
• AutoML I (virtual instructor-led mission)
• DataRobot for Data Scientists (virtual instructor-led mission)

Technical requirements

• Chrome browser
• DataRobot Automated Machine Learning — If you don’t have access to the application, please sign up for our free trial: datarobot.com/trial.