Exploratory Data Analysis Study Guide — MLS-C01

Exploratory Data Analysis (24% of Exam)

This domain tests your ability to select the right AWS service and technique for cleaning, transforming, and understanding datasets before model training. Expect scenario-based questions that ask you to choose between services (Data Wrangler vs. Glue DataBrew vs. Athena), justify encoding or scaling choices given algorithm constraints, and identify the correct remediation for data quality issues like missing values, outliers, and class imbalance.

Amazon SageMaker Studio and Data Wrangler

SageMaker Studio is the primary environment for interactive EDA. Notebooks run on managed compute instances and integrate directly with S3, Feature Store, and other SageMaker components. Use Studio when you need custom code-based analysis or visualizations beyond what no-code tools offer.

SageMaker Data Wrangler is the no-code/low-code answer for data preparation. Key exam facts:

• Connects to S3, Athena, Redshift, EMR, Snowflake, and Databricks
• Generates a Data Quality and Insights Report automatically — covers missing values, class imbalance, feature correlation, and target leakage detection
• Supports 300+ built-in transforms (the exam sometimes cites 300, sometimes 250+; either signals Data Wrangler)
• Outputs a processing job script you can export to SageMaker Processing, Pipelines, or a Jupyter notebook
• Decision rule: When the question mentions "no-code data prep," "feature transforms without writing code," or "data quality report," the answer is Data Wrangler, not Glue DataBrew

AWS Glue DataBrew

DataBrew is a visual data preparation tool that lives outside SageMaker. Key distinctions:

• 250+ transforms — the exam uses this number specifically for DataBrew (vs. 300+ for Data Wrangler)
• Works on data in S3, Glue Data Catalog, Redshift, RDS, and other JDBC sources
• Produces data profiling reports with column-level statistics (min, max, cardinality, missing %)
• Supports data quality rules you define to flag or fail a job based on thresholds
• Outputs cleaned datasets back to S3 or Glue
• Decision rule: Choose DataBrew when the question involves data engineers working outside the ML pipeline, catalog-integrated workflows, or the specific "250+ transforms" phrasing

Amazon Athena for EDA

Athena runs serverless SQL against data in S3. Use it for large-scale EDA when you need aggregations, distributions, or joins across raw data without loading it into a database.

• Integrates with Glue Data Catalog for schema management
• Supports partitioning to reduce query costs — exam may ask how to optimize Athena costs (answer: partition by date or high-cardinality filter columns)
• Common EDA queries: COUNT(*), GROUP BY, PERCENTILE_CONT, CORR for correlation
• Trap: Athena is for exploration and querying, not for training directly. You still need SageMaker Processing or Data Wrangler to feed features into training jobs

Amazon QuickSight

QuickSight is the BI layer for visual exploration. Exam focus is narrow:

• Use for dashboards and visualizations shared with business stakeholders
• SPICE is the in-memory engine — improves dashboard performance but has a storage limit (10 GB per user on standard; not usually tested precisely)
• Connects to Athena, S3, Redshift, RDS, and SageMaker outputs
• Decision rule: If a question asks about sharing interactive visualizations with non-technical users or building BI dashboards, the answer is QuickSight, not SageMaker Studio