Akademik Veri Yönetim Sistemi
JOURNAL OF SUPERCOMPUTING, cilt.81, sa.15, 2025 (SCI-Expanded, Scopus)
As dependency of computer network topologies on the internet increases, they turn into more defenseless to cyber-attacks. In extensive cyberspace, for detecting anomalous attacks and malicious activities, network Intrusion Detection Systems (IDS) are the crucial defense actions in current networks. For playing a central role in network environment, Deep Learning (DL) approach has been recently applied in designing intrusion detection systems. In this work, a hybrid deep learning model (CNN-LSTM) combining Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) is proposed to build a network IDS for both of the binary and multi classifications. For improving the detection performance of the CNN-LSTM model, swarm intelligence-based algorithms such as Particle Swarm Optimization (PSO) and Salp Swarm Algorithm (SSA) are used to optimize the hyperparameters of the proposed model. In the training and testing process, the CIC-IDS 2017 dataset is utilized for tuning the hyperparameters and verifying the proposed model. Furthermore, the CNN and LSTM are separately trained using the same optimization algorithms on the same dataset and also examined for comparing the proposed CNN-LSTM in terms of effective intrusion detection. Experiments have demonstrated that the proposed model yields more effective and robust intrusion detection performance by achieving remarkable results. Specifically, for the binary classification task, the CNN-LSTM model optimized by SSA achieved an accuracy of 99.83%, precision of 99.41%, recall of 99.58%, and F1-score of 99.49%. For the multi-class classification task, the SSA-optimized CNN-LSTM model obtained an accuracy of 99.7%, precision of 97.3%, recall of 99.7%, and F1-score of 99.7%. These results clearly indicate the superior detection capability and improved efficiency of the proposed model.