Locking Decision Tree with State Permutation Obfuscation: Software Implementation

This paper presents a mechanism for enhancing the security of decision tree machine learning models by employing finite state machine (FSM) permutation obfuscation. Our approach obscures the internal structure of the decision tree through key-driven state transitions, thwarting attackers from deciphering the logic or extracting sensitive information. We demonstrate the effectiveness of our method with a Python-based software prototype using the MNIST dataset, maintaining accuracy while deterring several attacks including brute-force, side-channel, and reverse engineering attempts. The implementation lays the foundation for replicating the mechanism on hardware platforms like FPG As, enabling efficient and secure deployment in resource-constrained environments.