Data Prep and Target Leakage in H2O Driverless AI
- Predictive AI
- H2O Driverless AI
This lesson walks through preparing a dataset for H2O Driverless AI using a loan default example, covering what makes data ready for training versus what still needs cleanup. It shows the Live Code option for writing custom Python and pandas transformations directly in Driverless AI, including dropping missing values, converting column types, sorting, and creating new columns, then saving the result as a reusable data recipe. The lesson also covers the difference between the training dataset and the scoring dataset in size and frequency of use, and explains target leakage, using a loan default reason column as an example.
Key takeaways
- How to use the Live Code option in Driverless AI to apply pandas transformations directly to a dataset
- What common data quality issues Driverless AI can handle automatically, like missing values and format errors
- Why training datasets need far more rows than the scoring dataset, which can run on a single row
- What target leakage is and why a column like a loan default reason should be dropped before training
- How to save a data prep recipe so it can be reused or shared across future datasets
Start owning your AI