Mondrian K-Anonymity

This implementation (following the paper) supports numerical, date and categorical attributes: each categorical attribute must have its own generalization hierarchy.

Usage

There must be a folder containing the .csv file with the data to anonymize and the .csv file which specifies the type for each attribute. All the results will be saved in a folder named Results.

This is an example of how you can use the program.

 python3 anonymizer.py

In this case Mondrian is executed with all the default settings: synthetic dataset provided by us and K=10.

Parameters

• -folder_name: name of the folder of the data (default: Dataset_synthetic)
• -dataset_name: name of the dataset file. (default: mainDB_10000.csv)
• -columns_type: name of the file containing the types (default: columns_type.csv)
• -K: the integer K (defualt: 10)
• -result_name: name of the file that will contains the data anonymized (defualt: anonymized.csv)
• -save_info: boolean value indicating either save in a txt file some info about the resulting data or not save any .txt file (default: True)

Here a further example using the parameters

python3 anonymizer.py -folder_name=Dataset_synthetic -dataset_name=mainDB_500000.csv -K=100 -result_name=anonymized_500000.csv -save_info=False

Requirements

pip3 install -r requirements.txt

• numpy==1.18.5
• pandas==1.1.2
• argparse==1.4.0
• datetime==4.3
• matplotlib~=3.2.1

Data

We provide 2 kind of data as an example: one synthetic and one real.

Synthetic

Each record has the following attributes:

• Gender: male or female
• Age
• Zipcode: code of the place where the person lives
• B-City: name of the city where the person was born
• B-day: birthday date
• Start therapy: date starting the therapy
• End therapy: date end treatment
• Blood type
• Weight (Kg)
• Height (cm)
• Disease: kind of disease affecting the person, attribute considered as Sensitive data

The records are generated randomly by means of a function provided in the project.

Real

This is taken from here. The records containing missing values are removed and we get rid of education-num and final-weight attributes. The annual-gain is treated as Sensitive data.

Contributors:

• Alessandro Caroti
• Simone Campisi
• Jacopo Dapueto