Data Preparation for Machine Learning – MLA-C01 Study Guide
This is the largest domain (28% of the exam) and reflects a real truth of ML engineering: most of the work is data. It covers ingesting data, choosing storage formats, cleaning, feature engineering, splitting, handling imbalance and bias, and the AWS services that support each step. Questions are scenario-based and service-aware — know which tool does what, and the standard techniques for common data problems.
Ingestion and Storage
- Batch ingestion — data lands in Amazon S3 (the ML data lake), often via AWS Glue ETL jobs.
- Streaming ingestion — high-velocity events (clickstream, IoT) via Amazon Kinesis Data Streams for near-real-time processing.
- Storage format — prefer compressed, columnar Parquet with sensible S3 partitioning for efficient reads and lower cost, versus large uncompressed CSV or millions of tiny objects.
"Continuous high-throughput events" → Kinesis. "Serverless ETL with a catalog and Spark" → AWS Glue.
Cleaning Data
Common issues and standard handling:
| Issue | Handling |
|---|---|
| Missing values | Impute (mean/median or model-based) or drop rows/columns based on impact |
| Outliers | Investigate; cap, transform, or remove if erroneous |
| Sentinel error codes (e.g., -9999) | Treat as invalid/missing — they're not real measurements |
| Duplicates | Remove to avoid biasing the model |
Sign in free to read the full guide
4 more sections (about 318 words) continue below. The first domain's study guide is free for every MLA-C01 learner with a free account.