Dependencies

Python3 and some python3 libraries:

• numpy (v1.15.4 used)
• scipy (v1.1.0 used)
• sklearn (v0.20.0 used)
• imblearn (v0.4.3 used) https://github.com/scikit-learn-contrib/imbalanced-learn
• matplotlib (to plot, v3.0.3 used)

Content

folder 'datasets':

• contain the 22 datasets used in the experiments

folder 'toy_example':

• contain a script to generate the figure describing the constraints

folder 'experiments_a_b':

• contain the source code to reproduce experiments a and b from the paper: without data processing (a) and with SMOTE and Random Undersampling (b)
• the experiment is launched with 'main.py' which apply the algorithms on the datasets, and store the results in a subfolder
• the file 'latex.py' load the result files, perform a mean over the files and generate latex code producing an array of the mean results

folder 'experiments_c_d':

• contain the source code to reproduce experiments c and d from the paper where we artificially increase the imbalance in the datasets
• the experiment is launched with 'main.py' which apply the algorithms on the datasets, and store the results in a subfolder
• the files 'plotAccuracyF1.py' and 'plotF1.py' load the results files, perform a mean over the files and generate PDF images of the results