suda 0.2.0

Python library for implementing Special Uniques Detection Algorithm (SUDA) for measuring disclosure control risk in synthetic data

suda

Sample uniqueness scoring in Python

This is a Python library for computing sample uniques scoring using the Special Uniques Detection Algorithm (SUDA).

The algorithm looks for rows in a dataset which are unique with respect to a number of category fields and scores them according to risk.

The smaller the number of fields for which a row is unique, the higher the score. So a row which has a unique value for a single field will score highly.

The more combinations by which a row is unique the higher the score. So a row which is unique in multiple ways will score highly.

Usage

Python

Call the suda() method with the dataframe to score, the maximum MSU to test for, the DIS score for the file (defaults to 0.1) and the columns to use for scoring (defaults to all columns).

For example, calling:

results = suda(data, max_msu=2)

Will score the 'data' dataframe and find MSUs of up to two fields. If the dataframe contained fields 'gender', 'age', 'education' and 'employment' then the algorithm will look for rows that are unique for all combinations of one and two fields (gender, age, education, employment, gender & age, gender & education, gender & employment, age & education, age & employment, education & employment.)

The output may look like:

id	msu	suda	fK	fM	gender	region	education	employment	dis-suda
0	0.0	0.0	2.0	0.0	female	urban	secondary incomplete	employed	0.000000
1	0.0	0.0	2.0	0.0	female	urban	secondary incomplete	employed	0.000000
2	1.0	12.0	1.0	4.0	female	urban	primary incomplete	non-LF	0.020690
3	0.0	0.0	2.0	0.0	male	urban	secondary complete	employed	0.000000
4	1.0	16.0	1.0	6.0	female	rural	secondary complete	unemployed	0.027586
5	0.0	0.0	2.0	0.0	male	urban	secondary complete	employed	0.000000

fK is the minimum frequency of the row - if this is >1 then there are no sample unique values for the row.

fM is the number of MSUs found for the row.

msu is the Minimum Sample Unique for the row - that is, the smallest number of fields where the row is unique.

suda is the SUDA calculated score, adding together the individual MSU scores (each MSU score is the factorial of the number of attributes in the dataset minus the MSU.)

dis-suda is the file-level risk score (DIS) divided by the total SUDA scores, multiplied by SUDA for the row. In other words, the total risk distributed by the rows.

Command line

Use the command line function to supply a CSV file for the input, a path to output the resulting CSV, the minimum MSU, the columns to include, and the file-level risk (DIS).

References

Elliot, M. J., Manning, A. M., & Ford, R. W. (2002). A Computational Algorithm for Handling the Special Uniques Problem. International Journal of Uncertainty, Fuzziness and Knowledge Based System , 10 (5), 493-509.

Elliot, M. J., Manning, A., Mayes, K., Gurd, J., & Bane, M. (2005). SUDA: A Program for Detecting Special Uniques. Joint UNECE/Eurostat Work Session on Statistical Data Confidentiality. Geneva.