arkhn-arx 0.0.8

arkhn_arx is a tool to pseudonymize or anonymize datasets while evaluating reidentification risk metrics

arkhn_arx

arkhn_arx is a module for dataset pseudonymization or anonymization which wraps pyarxaas

Install

pip install arkhn_arx

Connection to ARXaas service

This module uses https://github.com/navikt/arxaas service.

To run this service locally :

1. Make sure Docker Desktop is running
2. Pull the Docker image

docker pull navikt/arxaas

3. Run the Docker image

docker run -p 8080:8080 navikt/arxaas

Anonymization

Principle

This module can be used in 3 modes : to evaluate reidentification risk of a dataset, pseudonymize dataset or anonymize dataset. Anonymization is performed using k-anonymity and l-diversity algorithms.

• k-anonymity ensures that the information for each person contained in the release cannot be distinguished from at least k-1 individuals whose information also appears in the release (defining a k-anonymity group).
• l-diversity ensures that sensitive attributes are well represented (at least l distinct values) in each k-anonymity group

Arguments

• input_dataframe to anonymize
• configuration_file : json file containing anonymization parameters

config_dict = {"anonymization":{"type": 2, "k":2, "l":2},
                "attributes":[
                            {"customName":"att_1",
                             "att_type":"att_type"
                             "hierarchy_type":"hierarchy_type"}, 
                            ]
                }

• Anonymization parameters:

  • type : 0 returns risk metrics for initial dataset, 1 pseudonymize dataset, 2 anonymize dataset
  • k : parameter for K-anonymity
  • l : parameter for l-diversity

• Attributes parameters: for each attribute gives :

  • customName : column name of attribute in dataframe
  • att_type : attribute type for anonymization, can be:
    • "insensitive" : will be kept unmodified
    • "sensitive" : will be kept as-is but they can be protected using privacy models, such as t-closeness or l-diversity
    • "quasiidentifying" : will be transformed using hierarchies
    • "identifying"" : will be removed from the dataset
  • hierarchy_type : type of hierarchy to apply to attribute for anonymization, can be:
    • "interval" : can be used for variables with a ratio scale, intervals are defined using attribute quantiles
    • "date" : can be used for dates
    • "redaction" : can be used for a broad spectrum of attributes, masking parts of variables
    • "order" : NOT IMPLEMENTED can be used for variables with an ordinal scale, defining ordered group of variables

• URL_link to ARXaaS service : if ARXaas service is running locally URL is : "http://localhost:8080"

Example

You can test this module using the example.py script