Malicious Traffic Detection in Multi-Environment Network Using Dual-Data Trained LightGBM Approach

Cybersecurity is gaining interest from researchers as it becomes increasingly important to protect data communication on the internet from cybercriminals. However, with advancements in network infrastructures and architectures, network technology presents challenges in developing security systems that can effectively counter cyberattacks with high accuracy and low computational cost. The latest network architectures necessitate a system capable of analyzing diverse network traffic. To address this, our proposed approach targets the detection of malicious traffic in a multi-environment (M-En) network that includes SDN, IoT, and traditional IP-based networks. To tackle the diversity of M-En networks, we introduce the Dual-Data Trained Light Gradient Boosting Machine (DDT-LightGBM), which combines two LightGBM models through soft voting. The training strategy involves one LightGBM model trained on balanced data using SVMSMOTE for data balancing and another trained on an imbalanced dataset to inject diversity into individual training. By merging both models through soft voting and averaging probabilities for final predictions, the DDT-LightGBM approach achieves a significant accuracy score of 0.992, surpassing other methods in both accuracy and efficiency.