A Machine Learning-Enhanced Cybersecurity and Optimization Framework for Intelligent Threat Detection and System Efficiency

Modern Security should be context-aware is required to adapt to current Hyper-connected world, where Advanced and on-the-fly developed threats are emerging. In this paper we introduce MLECOF: a Machine Learning Enhanced, Cybersecurity & Optimization Framework that proposes a complete end to end horizontally scalable approach for real time threat detection, as well as the closed loop to optimize the infrastructure by detecting anomalies. MLECOF comprises of three major modules referreded as Data Ingestion and Preprocessing Module, Threat Detection and Classification Engine and Resource optimization and response unit (RORU). Both unsupervised and supervised machine learning: Autoencoder, Random Forest, XGBoost, CNN (Convolutional Neural Network), are ensemble under one framework to handily making robust distinction among the different threats with ultra-high accuracy. Then MLECOF is tested on popular datasets (CICIDS2017) and achieves the higher malicious accuracy (98.7% at best) with the aid of MOGA (Multi-Objective Genetic Algorithm) on system level performance optimization. The results outperform dramatic reductions in CPU and response latency and energy consumption, which verify that MLECOF can be deployed as secure, smart and efficient cyber defense solution in cloud, edge, and hybrid environment.