Profile before changing anything
Start with missing values, detected types, and distributions. A blank field may mean unknown, not applicable, or a collection error. Understanding that distinction helps you choose an appropriate treatment rather than applying the same rule to every column.
Review suggested fixes
The cleaning workspace flags issues and offers supported operations for you to review. Apply changes that fit your analysis and check the results. Automated suggestions help you find candidates for cleanup; they cannot decide what is correct for every business context.
Example: prepare data for a predictive model
For a customer dataset, examine incomplete numeric fields and inconsistent categories before engineering features. Decide how missing information should be represented, review unusual values against the source, and export the prepared data. When building a model, keep evaluation data separate and avoid using information that would not be available at prediction time.
Keep a usable output
Export a cleaned file for another tool, a report for review, or a notebook for further work. Preserve your original source data and compare the output with the original so that you can explain the preparation choices you made.
How to get started
- Profile the original dataset and identify quality issues.
- Review and apply suitable cleaning suggestions.
- Inspect the result, engineer features, and export.