100+ Free DataRobot Certified Professional Practice Questions

Automated Machine Learning (AutoML) concepts and workflow

Understand Autopilot modes (Quick, full, Manual, and Comprehensive), how blueprints and the Repository define modeling pipelines, the staged sample-size approach (16/32/64%), blender (ensemble) models, and the recommendation process that prepares the most accurate non-blender model for deployment.

Data ingestion, EDA, and feature engineering

Work through EDA1 summary statistics and EDA2 target analysis, ACE-based feature importance, target-leakage detection with the 0.85 moderate and 0.975 high-risk thresholds, Informative Features and Leakage-Removed lists, missing-value imputation, interaction transformations, and Feature Discovery across secondary datasets with a defined prediction point.

Model building, leaderboard evaluation, and selection

Interpret the Leaderboard ranked by the optimization metric (LogLoss for classification; RMSE, Tweedie, Poisson, or Gamma Deviance for regression), the default 20% holdout with five-fold cross-validation, ROC curve and threshold tools, Lift and Cumulative Gain charts, asterisked in-sample scores, and the Recommended and Prepared for Deployment badges.

Model interpretation

Use Feature Impact (default permutation importance) for global drivers, Feature Effects to see how feature values shape predictions, and Prediction Explanations via SHAP (additive, signed contributions) or XEMP (works for all models after Feature Impact is computed), including text-level explanations, to justify individual decisions.

Deployment, predictions, and MLOps monitoring

Register and deploy models through the Model Registry, choose real-time versus scheduled batch predictions, and monitor the three deployment health indicators: service health (latency, throughput, errors), data drift (target drift against the holdout baseline and feature drift), and accuracy (using actuals matched by association ID).

Applying DataRobot to business problems

Frame the business question and target, set the prediction point to avoid leakage, translate ranked predictions into action with Lift and Profit curves, estimate ROI from payoff matrices, and sustain trusted, compliant AI by retraining when drift or accuracy decline across the model lifecycle.