Customer records (cleaned)

What it is good for

• Verify a cleaning pipeline by comparing your output with a reference result.
• Quantify how cleaning changes summaries (row count, mean spend, missingness).
• Discuss which cleaning steps are rules and which are judgement calls.

Methods: data validation, missing-data summaries, before/after comparison.

Data quality notes

• 679 rows (the 21 exact duplicates removed).
• Nulls: age 23 (18 originally missing + 5 impossible), city 7, annual_spend_sar 16 (12 missing + 4 negative), referral_source 28 (placeholders).
• 6 halala entries converted to SAR; 45 city spellings standardised; 8 referral values trimmed.
• Remaining missing values are genuine unknowns; they were not imputed.

Provenance and licence

Generated by running the customers_messy generator and then applying, in order: (1) drop exact duplicate rows, keeping the first occurrence; (2) city: trim spaces, lower-case and map spellings and aliases to one canonical name (Riyadh, Jeddah, Dammam, Khobar, Abha); (3) referral_source: trim spaces and turn placeholders "N/A", "unknown", "-" (any case) into null; (4) age outside 16–110 → null; (5) negative annual_spend_sar → null; (6) annual_spend_sar above 100,000 → divided by 100 (halalas to SAR).

Synthetic data · CC0-1.0 (synthetic data generated by Statistics Quest) · generated deterministically (seed 20241025). It does not describe real people or organisations; any resemblance is coincidental.