Machine-Learning Ciphertext Decryption Algorithm

This project introduces a decryption tool that uses a multi-class SVM classification model to decrypt ciphertexts, encrypted with some randomly generated mixed-ciphertext alphabet.

• Implement manual feature extraction: Identifies and describes the most common features that define the internal structure of the text-datasets (training & testing). These features include: Single Letter Frequencies, Letter Occurencies in k-letter words, Letter Position Frequencies and Double Letters Frequencies.
• Perform manual feature selection: Creates feature-set (X) and label-set (y), by selecting the features that describe best each class.
• Implement the classification model iteratively: Trains an SVM classifier on the training plaintext. It then uses this classification model iteratively, to assign class-labels to the testing ciphertext (decryption alphabet prediction).
• Decrypt the testing ciphertext: Applies the predicted decryption alphabet to the testing ciphertext to decrypt it.

Take a look at this demo code in NBViwer :)

Prerequisites

The following python packages are required for the code to run:

• Python 3: https://www.python.org/downloads/
• NumPy: pip install numpy
• Scikit-learn: pip install -U scikit-learn
• Matplotlib: pip install numpy sklearn matplotlib

Alternatively: you can download requirements.txt and run pip install -r requirements.txt, to automatically install all the packages needed to reproduce my project on your own machine.

The code uses the TRAINING-tolstoy-anna-karenina.txt and TESTING-pushkin-eugene-onegin.txt files as the training and testing text. Make sure that these files are in the same directory as the code.

Acknowledgments

I would like to express my gratitude to Interactive Maths for providing valuable information and resources, that contributed to the development of my project.

All of the books used as training & testing datasets in this project, were obtained from Project Gutenberg. Therefore, I would like to acknowledge its invaluable contribution in making these texts freely available for research and analysis.

Conclusion

This code provides a basic implementation of an ML Ciphertext Decryption Algorithm. Users are encouraged to modify the training/testing datasets or the feature-tuple, to observe the impact on the total performance and accuracy.

Here are some suggestions:

# Remove some good features from the feature tuple:
117  fig, axs = plt.subplots(nrows=3, ncols=3, figsize=(12, 8))
...
120  for d in (f0, f1, f2, f3, f4, f8, f9, f10, f11):

# Use a different testing dataset:
87   training_text = "TRAINING-tolstoy-anna-karenina.txt"
88   testing_text = "TESTING-goethe-werther.txt"
89   decryption_alphabet = "ghbcafmsztwnroevlixupjyqkd"  # encryption_alphabet = "ecdzofabrvyqglnuxmhjtpkswi"

_{In this folder, you will find a variety of texts to use for the testing process. You can also create another training dataset from scratch, however it requires a lot of effort and is not recommended.}

Author

Natalia Koliou: find me on LinkedIn.