Data Cleaning & Preprocessing for Data Analysts

Most of an analyst's time goes into cleaning, not analyzing, and this course treats that seriously. Five compact modules cover the full workflow: inspecting and typing a new file, handling missing data and duplicates, standardizing text and catching invalid values, merging and reshaping multiple sources, and finally building one reusable, validated cleaning pipeline. Every lesson runs on a realistically messy dataset with genuine data quality problems built in on purpose, so what you're fixing is real, not simulated. Closes with two portfolio projects: a multi-source merge and a full reshape-and-validate pipeline.

Rutvik Acharya

Principal Data Scientist

Atlassian

Loading modules...

What You'll Learn

Follow a repeatable Inspect, Profile, Clean, Validate, Document workflow on any new dataset

Fix parsing issues, missing data, and duplicate records without silently corrupting your analysis

Standardize inconsistent text and catch invalid values using business rules, not guesswork

Merge multiple real-world data sources and reshape data correctly, validating that nothing broke

Who Should Attend

Data Analysts who spend more time fighting messy data than analyzing it

Anyone who has shipped a wrong number because of a silent join or parsing bug

Analysts moving from spreadsheets to Python/pandas who need a real cleaning workflow

Teams standardizing how they validate data before it reaches a dashboard or report

CERTIFICATION

Certificate of Completion

6 Modules

4 Hour 20 Minutes

12 Lessons

12 Challenges

Language: English

FAQ

FREQUENTLY ASKED QUESTIONS

Real problems generated on purpose: dates stored in two different formats in the same column, prices stored as text with currency symbols, duplicate records, and one column with 13 different spellings of what should be 4 categories — not a sanitized version of messiness.

By checking whether the missingness is scattered or clustered in a specific group first. A value missing at random is handled differently than a gap that's concentrated in one store or one time period — the course teaches you to check before deciding.

It distinguishes exact duplicates, which are generally safe to remove, from duplicate keys with conflicting data, which are real data integrity problems that need investigation rather than automatic deletion. This includes a real example of a duplicate customer ID that would have silently multiplied a merge's row count if left unresolved.

Yes — for merging records that don't share a clean ID, like matching an email marketing list to a CRM by name alone. The portfolio project includes a real case where 95% of customers share a name with someone else, making name-only matching unsafe without a second signal.

By checking the row count before and after against what the join logic predicts, and explicitly checking for orphaned keys — the same discipline used throughout the course, applied specifically to the moment joins most often go wrong silently.

A single function that runs the full cleaning workflow and produces three things automatically every time: pass/fail validation checks, a before-and-after summary of the data's shape, and a plain-language log of what was found and changed.

It explains why the shape matters — a wide format can't support a groupby or a join the way long format can — then shows the reshape and validates the row count arithmetically afterward to catch a broken reshape immediately.

Through domain rules based on what's actually possible in the business — such as a negative age or a future order date — rather than statistical thresholds like IQR fences. The two approaches catch different problems, and the course is explicit about which one this is.

A merge where a duplicate key in a reference table silently multiplies rows in the result — the course walks through a case where this exact thing happens and shows how to catch it before it corrupts a downstream report.

Yes — a channel column with 13 spellings of 4 real payment methods is shown producing a wrong-looking breakdown until standardized, directly demonstrating why an analysis can only be correct when cleaning happens first.

Scroll Up

GET FREE CAREER COUNSELING