An Intelligent Machine Learning Framework for Encryption Algorithm Selection Based on Performance Prediction

Abstract

Selecting an appropriate encryption algorithm is a critical challenge in modern cyber security, as different algorithms exhibit varying levels of computational efficiency and security depending on file characteristics. This study presents an intelligent framework for encryption algorithm selection based on experimental performance evaluation and Machine Learning prediction. Four encryption algorithms, namely AES-256, DES, Blowfish, and Hybrid RSA-2048 + AES-256, were experimentally evaluated using multiple file types and file sizes. Performance was assessed using encryption time, decryption time, CPU utilization, RAM consumption, throughput, and data integrity. The resulting experimental dataset was then used to train and evaluate four regression models: Linear Regression, Decision Tree Regression, Random Forest Regression, and Gradient Boosting Regression. Comparative analysis demonstrated that Random Forest Regression achieved the highest predictive performance among the evaluated models. Based on these findings, a Smart Encryption Recommendation Framework was developed to predict encryption performance and recommend the most appropriate encryption algorithm according to file characteristics before the encryption process is executed. The proposed framework combines experimental cryptographic evaluation with Machine Learning to support intelligent decision-making, improve encryption algorithm selection, and enhance computational efficiency while maintaining reliable data protection. The novelty of this work lies in introducing a predictive decision-support framework that recommends the most appropriate encryption algorithm before execution based on file characteristics and predicted computational performance, thereby extending conventional encryption benchmarking toward intelligent encryption decision support.