Analysis of an anonymization method

After the creation of a method to anonymize time series (TS), it is important to compare orginial and anonymized TS. Here a four scripts to :

1. measure normalized multivariate dissimilarities,
2. generate CSV files with average, standard deviation, median, minimun or maximum values from the distribution of normalized dissimilarities,
3. generate CSV with statistical test results for one physiological parameter and a pair of parameters,
4. make a boxplot illustating normalized dissimilarity distributions between original and anonymized TS for all parameters.

analysis_anonym_meth1_part1.py

This python script allows to measure normalized multivariate dissimilarities between real patient timeseries and anonymized timeseries.

Precondition :

To have the following packages installed : sys, os, re, glob, time, random, statistics, csv, pandas, array, numpy, dtaidistance. If necessary, please look at the install.txt file in the root folder.

Execution

Command line: python3 analysis_anonym_meth1_part1.py path_pr path_pa path_analysis_anonym filename_csv

The differents arguments:

1. path_pr corresponds to the path of the folder containing original multivariate timeseries.
2. path_pa corresponds to the path of the folder containing anonymized multivariate timeseries.
3. path_analysis_anonym corresponds to the folder in which the pdf graphical timeserie file will be saved. The script creates the directory if it doesn't exist.
4. filename_csv corresponds to the output filename of normalized dissimilarities.

Example command line (from the root folder): "python3 tools/analysis_anonym_meth1_part1.py multivariate_original_dataset/ gener_simulated_data_meth1_prop-level_0.50_perturb-level_0.05/ analysis_anonym_meth1_prop-level_0.50_perturb-level_0.05/ distri_dissim_norm_meth1_prop-level_0.50_perturb-level_0.05.csv". This command line will create statistical datasets for each physiological parameter.

analysis_anonym_meth1_part2.py

This python script allows to generate the different csv files containing either average, standard deviation, median, minimun or maximum values from the distribution of normalized dissimilarities.

Precondition :

All packages cited in the analysis_anonym_meth1_part1.py script have been installed and are commun to both scripts (part1 and part2).

Execution

Command line: python3 analysis_anonym_meth1_part2.py path_pr path_pa path_analysis_anonym val_prop val_perturb

The differents arguments:

1. path_pr corresponds to the path of the folder containing original multivariate timeseries.
2. path_pa corresponds to the path of the folder containing anonymized multivariate timeseries.
3. path_analysis_anonym corresponds to the path of the folder containing new csv files (e.g., dissim_norm files) and where the new analyses will be saved in.
4. val_prop corresponds to the proportion level that was applied to measure dissimilarities.
5. val_perturb corresponds to the perturbation level that was applied to measure dissimilarities.

Example command line (from the root folder): "python3 ./tools/analysis_anonym_meth1_part2.py ./multivariate_original_dataset/ ./gener_simulated_data_meth1_prop-level_0.50_perturb-level_0.05/ ./analysis_anonym_meth1_prop-level_0.50_perturb-level_0.05/ 0.50 0.05". This command line will create csv files containing statistical values (avg, std, med, min & max) for 0.50 proportion and 0.05 perturbation levels.

analysis_anonym_meth1_part3.py

This python script allows to generate the different csv files containing statistical test results for one physiological parameter and a pair of parameters.

Precondition :

To have the following packages installed : itertools, scipy. If necessary, please look at the install.txt file in the root folder.

Note that some packages used in this script (analysis_anonym_meth1_part3) have already been installed for the two previous scripts (part1 and part2).

Execution

Command line: python3 analysis_anonym_meth1_part3.py path_analysis_anonym param_physio val_prop val_perturb

The differents arguments:

1. path_analysis_anonym corresponds to the path of the folder containing new csv files (e.g., dissim_norm csv files) and where the new analyses will be saved in.
2. param_physio corresponds to the physiological parameter that want to be analyzed (FC, PAS, PAM, PAD)
3. val_prop corresponds to the proportion level that was applied to measure dissimilarities (e.g., 0.50).
4. val_perturb corresponds to the perturbation level that was applied to measure dissimilarities (e.g., 0.05).

Example command line (from the root folder): "python3 ./tools/analysis_anonym_meth1_part3.py ./analysis_anonym_meth1_prop-level_0.50_perturb-level_0.05/ FC 0.50 0.05". This command line will create csv files containing statistical values (avg, std, med, min & max) for 0.50 proportion and 0.05 perturbation levels and only for the FC physiological parameter.

boxplot_meth1.R

This R script allows to generate and save the boxplot illustating normalized dissimilarity distributions between real and anonymized patients for all parameters (each pair of perturbation/proportion levels).

Precondition :

To have the following packages installed : readr, ggplot2. For their installation, please look at the install.txt file in the root folder.

Execution

Command line: Rscript ./tools/boxplot_meth1.R

This command line will save the generated boxplot into the ./tools/ directory as a .png file.