Dynamic selection of encryption algorithms with machine learning

This study aims to dynamically select the optimal encryption algorithm based on system parameters using machine learning (ML) classification algorithms. In the literature, it is observed that the algorithms used are generally selected manually. This manual selection leads to performance issues, especially in resource-constrained systems. However, in real-world applications, many parameters are decisive in algorithm selection. These include data size, encryption time, decryption time, memory usage, processor usage, energy consumption, and other characteristics. Accordingly, a 2000-line synthetic dataset consisting of 6 criteria, such as file size, encryption time, decryption time, processor and memory usage, and energy consumption, was generated using the Python programming language. Blowfish, Advanced Encryption Standard (AES), Rivest Cipher 6 (RC6), Serpent, and Kalyna were selected from symmetric encryption algorithms for the dataset. The data was tested in the model with seven different ML algorithms; Random Forests (RF), Extreme Gradient Boosting (XGBoost), and Histogram-Based Gradient Boosting (HistGB) models showed the highest success. The results show that the models performed consistently at a Macro-F1 level of ≈ 0.916 – 0.918. XGBoost was the fastest for inference time, RF was the fastest for training time, and HistGB was the highest for Macro-F1. In conclusion, the designed model proposes a structure focused on the needs of the system and aims to fill the gap in the literature in this context. The created architecture can be applied in the Internet of Things (IoT), mobile systems, and cloud computing environments, and the model's decision performance can be improved by using explainable artificial intelligence tools in the future.