Detection of Cipher Types Using Machine Learning Techniques

The identification of a cryptosystem has been a challenge for decades. This paper’s main objective is to identify the type of cryptosystem used to encrypt a particular text. We have explored the realm of machine learning to recognize a pattern among complex classical ciphertexts that generally have a simple representation in plaintext. We have modeled our objective as a sequence-to-sequence learning task that we have tried to solve using Convolution Neural Networks (CNNs) and state-of-the-art Transformer models. With only a tiny dataset (130 k) consisting of ciphertexts and the corresponding cryptosystem used to encrypt the same, our model has shown a good accuracy of 96.72 % which proves a significantly steep learning curve compared to other sequence-to-sequence models. Here we show the enormous potential of these models and how they can perform even better if the barrier of resources and computation time is lifted.