Data Profiling and Augmentation for AutoML
- Predictive AI
- Enterprise h2oGPTe
- H2O Driverless AI
This lesson walks through data profiling and augmentation for AutoML, starting with an overview of a customer churn dataset, including class imbalance, missing data, and feature types. It covers automatic quality checks that run within a Driverless AI experiment, such as leakage detection, ID column removal, and missing value imputation, alongside automatic feature engineering. It shows using h2oGPTe agents to ask questions about the dataset, surfacing data quality fixes and a recommendation to keep outliers since they may reappear in production. It closes by pushing the curated dataset directly into Driverless AI for modeling.
Key takeaways
- What AutoViz surfaces immediately about a dataset, including class imbalance, missing data, and feature types
- How Driverless AI automatically checks for leakage and drops ID columns before training a model
- Why an h2oGPTe agent might recommend keeping outliers rather than removing them before modeling
- How asking an h2oGPTe agent questions about a dataset can surface automatic data quality fixes
- The curated dataset can be downloaded or pushed directly into Driverless AI for model building
Start owning your AI