Key takeaways
- What automated machine learning actually automates across the pipeline: data visualization, data pre-processing, feature engineering, model selection, hyperparameter tuning, and documentation
- A live Driverless AI demo that turns a 14-column churn dataset into a deployable model in 15 minutes, generating 301 features and training 79 feature-tuning models plus 54 feature-evolution models along the way
- Why feature engineering is normally the most time-consuming step, and how automation handles techniques like target encoding, dimensionality reduction, and time series and text features
- How Driverless AI explains individual predictions and overall model behavior using Shapley values, partial dependence plots, and surrogate decision trees
- How autodoc automatically generates a Word-format model document for regulatory review and internal handoffs
- How H2O AI Hybrid Cloud extends AutoML into production deployment on Kubernetes, drift monitoring, and low-code app development for business users
Overview
In this session, H2O.ai's Lena breaks down what it actually takes to automate a machine learning pipeline, from data visualization and feature engineering through model training, explainability, and monitoring. Using Driverless AI, she runs a live churn demo that takes a 14-column dataset to a deployable model in 15 minutes, generating 301 engineered features along the way. The recording also covers automatic documentation, interpretability methods like Shapley values, and how H2O AI Hybrid Cloud extends AutoML into app development and operations for business users.