Dirty DataData Cleansing
XxxxxxDSCI 5240
December 4, 2012
Introduction
• Real data is dirty
• Why clean?–Eliminate duplicates–Smaller database–Accurate statistics
• The problem–Merge/Purge of large databases
Preview
• Data Cleansing Solutions
• Real World Data
• OCAR’s Data
• Conclusion
Data Cleansing Solutions
• Sorted-Neighborhood Method
• Equational Theory
• Transitive Closure
Sorted-Neighborhood Method
• Three phases– 1. create keys– 2. sort the data– 3. merge
• Three passes using different key– Multi-pass method
Sorted-Neighborhood Method
• Key selection
First Name Last Name Address ID Key
Sal Stolfo 123 First Street 45678987 STLSAL123FRST456
Sal Stolfo 123 First Street 45678987 STLSAL123FRST456
Sal Stolpho 123 First Street 45678987 STLSAL123FRST456
Sal Stiles 123 Forest Street 45654321 STLSAL123FRST456
Sorted-Neighborhood Method
• Sort using the key selected
First Name Last Name Address ID Key
Sal Stolfo 123 First Street 45678987 STLSAL123FRST456
Sal Stolfo 123 First Street 45678987 STLSAL123FRST456
Sal Stolpho 123 First Street 45678987 STLSAL123FRST456
Sal Stiles 123 Forest Street 45654321 STLSAL123FRST456
Sorted-Neighborhood Method
• A ‘window size’ is created for merging
First Name Last Name Address ID Key
Sal Stolfo 123 First Street 45678987 STLSAL123FRST456
Sal Stolfo 123 First Street 45678987 STLSAL123FRST456
Sal Stolpho 123 First Street 45678987 STLSAL123FRST456
Sal Stiles 123 Forest Street 45654321 STLSAL123FRST456
Merge Phase - Equational Theory
• A set of equation rules that defines equivalence
• A type of clustering function (pattern recognition)
• Rules may require an expert
Merge Phase - Equational TheoryEnglish rules:
Given two records, r1 and r2.
IF(the last names of r1 equals the last name of r2,
AND the first names differ slightly,
ANDthe address of r1 equals the address of r2)
THENR1 is equivalent to r2
Merge Phase - Equational TheoryResults
SSN Name (First, Initial, Last) Address
334600443 Lisa Boardman 144 Wars St.
334600443 Lisa Brown 144 Ward St.
525520001 Ramon Bonilla 38 Ward St.
525250001 Raymond Bonilla 38 Ward St.
0 Diana D. Ambrosion 40 Brik Church Av.
0 Diana A. Dambrosion 40 Brick Church Av.
0 Colette Johnen 600 113th St. apt.5a5
0 John Colette 600 113th St. ap. 585
850982319 Ivette A Keegan 23 Florida Av.
950982319 Yvette A Kegan 23 Florida St.
r1
r2
Merge Phase - Transitive Closure
• Applied to a single pass sorted-neighborhood method
• Improvement of accuracy
• Decreases processing time and cost
Merge Phase - Transitive Closure
English rules:
Given three records a, b and c.
IF (a is similar to b
ANDb is similar to c)
THENa is similar to c
Real World Data
• State of Washington Department of Social and Health Services
• Office of Children Administrative Research (OCAR) of the Department of Social and Health Services
OCAR’s Data• 6,000,000 records• Grows by 50,000 per month• 19 fields
– First and last name– Birthdate– SSN– Case number– Worker ID– Gender– Race– Service ID– Service dates– Payments
OCAR’s Data - Problems
• Names misspelled• Missing birthdates• Missing or wrong SSN• Multiple case numbers • Ghost records
OCAR’s Data - Goals
• To answer:– “How many children are in foster care?”– “How long do children stay in foster care?”– “How many different homes do children typically
stay in?”
OCAR’s Data - Cleaning• 128,438 records sampled (one service office)• Consulted with expert1
• 24 rules established• Used sorted-neighborhood multi-pass methods• Applied equational theory• Keys
– 1. Last name, First name, SSN, and Case number– 2. First name, Last name, SSN, and Case number– 3. Case number, First name, Last name, and SSN
1Timothy Clark, OCAR Computer Information Consultant
OCAR’s Data - Results
• Identified 8,504 individuals in sample
• 45.8% correctly classified
• 86.0% where correctly merged
• Multi-pass sorted-neighborhood confirmed
Review
• Multi-pass sorted-neighborhood method
• Equational method
• OCAR’s data
Conclusions
• Sort-neighborhood method can be expensive– During the sorting phase• Process time
• improved accuracy– Multiple times– Small windows– Computation of the transitive closure
Sources
• Real-world Data is Dirty: Data Cleansing and The Merge/Purge Problem; Mauricio A. Hernandez and Salvatore J. Stolfo; Department of Computer Science, Columbia University, New York, NY 10027.
• Haiguang Li, 2011 class presentation
• www.cs.columbia.edu/~sal
• http://www.dshs.wa.gov/default.shtm