Question

A data science team at an insurance company has a labeled tabular dataset in Amazon S3 and needs to predict whether a claim is fraudulent. They have limited ML expertise and want to quickly establish a strong baseline model, automatically exploring feature preprocessing, multiple algorithms, and hyperparameters, while still being able to inspect the generated candidate pipelines and code. Which approach best meets these requirements with the least manual effort?

Accepted Answer

Use Amazon SageMaker Autopilot to automatically build, train, and tune candidate models, then review the generated candidate notebooks and leaderboard.

Answer

Manually write an XGBoost training script and run a SageMaker automatic model tuning job with a Bayesian strategy over a large hyperparameter search space.

Answer

Deploy a pre-trained fraud detection model from SageMaker JumpStart directly to a real-time endpoint without additional training.

Answer

Use Amazon SageMaker Data Wrangler to build a feature transformation flow and export it, then hand off manual model selection to the team.