SageMaker Canvas Direct-to-Snowflake: XGBoost Fraud Model Feature Engineering
AWS ML Blog · rss · 2026-08-21
Part 2 of the series covers the hands-on core: connecting SageMaker Canvas directly to Snowflake, engineering features in Data Wrangler, and training an XGBoost fraud detection model.
- Connection setup only needs the Snowflake org-account ID and credentials, letting Canvas query the warehouse without manual exports.
- Feature design: card-level and merchant-level outlier thresholds via avg(amt)+3stddevpop(amt), plus transaction hour, day-of-week and age-bucket signals; two sources joined with a left outer join on Category, followed by dropping sensitive columns.
- Also walks through quick single-user Canvas domain creation.
More from Apps
- AI Security System Fail: Not Replacing Watchdogs Yet — AIandDesign · 2026-08-21
- Turning any livestream into a prediction market — moonsandhues · 2026-08-21
- First Impression of Grok Bot: Actually Useful Proactive Follow-up Messages — GregKamradt · 2026-08-21
- ChatGPT can now send texts for you with new Apple Messages plugin — TechCrunch AI · 2026-08-21
- Building an AI-native email client on Cloudflare Email, Workers and AI Gateway — ritakozlov · 2026-08-21
- Musk shares the easiest way to try Grok Build — a web link, no scripts or downloads — elonmusk · 2026-08-21