Enhancing Network Traffic Classification Using Multi-Tier Reinforced Salp Optimization Algorithm and Deep Learning Models

Abstract

Accurate and efficient detection of cyber threats is essential for effective network traffic analysis. This research introduces a Multi-Tier Reinforced Salp Optimization Algorithm (MTR-SOA) for feature selection, integrated with deep learning (DL) models to enhance network traffic classification. The proposed method is evaluated against traditional optimization techniques, including Particle Swarm Optimization (PSO), Grey Wolf Optimizer (GWO), and Genetic Algorithm (GA), across various DL models such as FNN, CNN, DELM, and LSTM. Experimental results reveal that MTR-SOA consistently delivers superior performance, with the LSTM model achieving the highest accuracy of 98.2% and demonstrating strong classification across all traffic categories. Furthermore, MTR-SOA reduces computational time by up to 30%, making it suitable for real-time network traffic analysis. Class-wise evaluation on the HIKARI-2021 dataset highlights its effectiveness in identifying complex cyber-attacks like XMRIGCC CryptoMiner and Probing. These findings confirm that integrating MTR-SOA with DL models enhances network traffic analysis by improving both accuracy and computational efficiency, offering a robust solution for detecting and classifying diverse traffic patterns.

Accurate and efficient detection of cyber threats is essential for effective network traffic analysis. This research introduces a Multi-Tier Reinforced Salp Optimization Algorithm (MTR-SOA) for feature selection, integrated with deep learning (DL) models to enhance network traffic classification. The proposed method is evaluated against traditional optimization techniques, including Particle Swarm Optimization (PSO), Grey Wolf Optimizer (GWO), and Genetic Algorithm (GA), across various DL models such as FNN, CNN, DELM, and LSTM. Experimental results reveal that MTR-SOA consistently delivers superior performance, with the LSTM model achieving the highest accuracy of 98.2% and demonstrating strong classification across all traffic categories. Furthermore, MTR-SOA reduces computational time by up to 30%, making it suitable for real-time network traffic analysis. Class-wise evaluation on the HIKARI-2021 dataset highlights its effectiveness in identifying complex cyber-attacks like XMRIGCC CryptoMiner and Probing. These findings confirm that integrating MTR-SOA with DL models enhances network traffic analysis by improving both accuracy and computational efficiency, offering a robust solution for detecting and classifying diverse traffic patterns.