Anonymizing categorical data

Data Privacy and Anonymization in Python

Rebeca Gonzalez

Instructor

Generalization

    Age    Gender    Department    Condition
0    30    F         Finance       Anxiety disorders
1    42    M         Production    Bronchitis
2    35    F         Marketing     Dysthymia
3    39    F         Production    Dysthymia
4    40    M         Marketing     Flu

    Age    Gender    Department    Condition
0    <40   F         Finance       Anxiety disorders
1    >=40  M         Production    Bronquitis
2    <40   F         Finance       Dysthymia
3    <40   F         Production    Dysthymia
4    >=40  M         Marketing     Flu
Data Privacy and Anonymization in Python

Generalization of categorical data

# See the dataset
hr.head()

    Age    BusinessTravel        Department                EducationField    EmployeeNumber
0    41    Travel_Rarely         Sales                     Life Sciences     1
1    49    Travel_Frequently     Research & Development    Life Sciences     2
2    37    Travel_Rarely         Research & Development    Other             4
3    33    Travel_Frequently     Research & Development    Life Sciences     5
4    27    Travel_Rarely         Research & Development    Medical           7
Data Privacy and Anonymization in Python

Categorical data

Limited or fixed number of possible values.

  • race
  • gender
  • hometown
  • age group
  • educational level
  • movies they like and preferences
Data Privacy and Anonymization in Python

Anonymizing categorical data

Histogram of the categorical values in Education Field

Data Privacy and Anonymization in Python

Anonymizing categorical data

     Department                EducationField
0    Sales                     Life Sciences
1    Research & Development    Life Sciences
2    Research & Development    Other
3    Research & Development    Life Sciences
4    Research & Development    Medical

Original dataset

     Department                EducationField
0    Sales                     Medical
1    Research & Development    Marketing
2    Research & Development    Life Sciences
3    Research & Development    Other
4    Research & Development    Life Sciences

Resulting dataset after sampling from the probability distribution of the educationField column in the original dataset.

Data Privacy and Anonymization in Python

Sample from data

The U.S. Census publicly releases samples of data that they collect about citizens.

Enable calculation of large-scale statistical patterns:

  • averages
  • variances
  • clusters
Data Privacy and Anonymization in Python

Explore the distribution

# Show the absolute frequencies of each unique value
hr['EducationField'].value_counts()

Life Sciences       606
Medical             464
Marketing           159
Technical Degree    132
Other                82
Human Resources      27
Name: EducationField, dtype: int64
Data Privacy and Anonymization in Python

Explore the distribution

# Generate a bar plot for the categories
df['BusinessTravel'].value_counts().plot(kind='bar')

Bar plot of the categorical values in the Business Travel column

Data Privacy and Anonymization in Python

Explore the distribution

# Obtain the absolute frequencies of each unique value
counts = hr['EducationField'].value_counts()

# Print the list of indexes
print(counts.index)

Index(['Life Sciences', 'Medical', 'Marketing', 
      'Technical Degree', 'Other', 'Human Resources'],
       dtype='object')
Data Privacy and Anonymization in Python

Explore the distribution

# Probability distributions of each unique value
counts = df['EducationField'].value_counts(normalize=True)

Life Sciences       0.412245
Medical             0.315646
Marketing           0.108163
Technical Degree    0.089796
Other               0.055782
Human Resources     0.018367
Name: EducationField, dtype: float64
Data Privacy and Anonymization in Python

Explore the distribution

# Values of the frequencies of each unique value
df['EducationField'].value_counts(normalize=True).values

array([0.4122449 , 0.31564626, 0.10816327, 0.08979592, 0.05578231,
       0.01836735])
Data Privacy and Anonymization in Python

Sampling from the same distribution

# Sample from a probability distribution
hr_sample['EducationField']= np.random.choice(counts.index, 
                                              p=counts.values, 
                                              size=len(hr))


# See resulting dataset
hr.head()

    Age    BusinessTravel        Department                EducationField    EmployeeNumber
0    41    Travel_Rarely         Sales                     Life Sciences     1
1    49    Travel_Frequently     Research & Development    Medical           2
2    37    Travel_Rarely         Research & Development    Marketing         4
3    33    Travel_Frequently     Research & Development    Technical Degree  5
4    27    Travel_Rarely         Research & Development    Medical           7
Data Privacy and Anonymization in Python

Sampling from the same distribution

# Show the absolute frequencies of each category
hr['EducationField'].value_counts()

Life Sciences       606
Medical             464
Marketing           159
Technical Degree    132
Other                82
Human Resources      27
Name: EducationField, dtype: int64

# Show the frequencies of the resulting column
hr_sample['EducationField'].value_counts()

Life Sciences       604
Medical             493
Marketing           158
Technical Degree    120
Other                61
Human Resources      34
Name: EducationField, dtype: int64
Data Privacy and Anonymization in Python

Let's practice!

Data Privacy and Anonymization in Python

Preparing Video For Download...