For more detail about the method refer to Curie A method for protecting SVM Classifier from Poisoning Attack

The required libraries are in requirements.txt. Install with: pip install -r requirements.txt

The files are in CSV format. Each row is one data point -- the last value is the class and the rest are the features.

Two examples datasets are included in data/.

artificial_20_2000.txt has 2000 data points, 20 features and 2 classes. artificial_20_2000_attack_50.txt is has 50 attack points inserted. artificial_20_2000_cleaned.txt is the cleaned file.

The attack data were genererated with AdversariaLib

python3 filter.py <source_file> <dest_file>

<source_file> is the file that contains the attack point. <dest_file> is the file where the filtered data will be saved.