Network Intrusion Detection Using Symbiotic Organism Search and Deep Learning

Abstract

With the increasing complexity of network environments and the rise in cyber threats, effective and accurate network traffic classification has become crucial for maintaining security and efficiency. Traditional classification methods face challenges in handling large-scale data and class imbalances. This research proposes a novel hybrid approach that integrates the Symbiotic Organisms Search (SOS) algorithm with a Convolutional Neural Network (CNN) to enhance classification performance in flow-based intrusion detection systems. The model utilizes statistical features extracted from TCP flows specifically packets and bytes captured via Wireshark and preprocessed as time-series data. The SOS algorithm is employed to optimize the feature set before feeding it into CNN for classification. Three datasets (USTC-TFC2016, ISCX VPN- nonVPN, and LBNL/ICSI) were used to validate the proposed method. Initial results using only CNN achieved a classification accuracy of 91.27%, but suffered from class imbalance issues, particularly in identifying normal traffic (class 0). The integration of the SOS algorithm significantly improved the model’s performance.  The final model achieved a remarkable accuracy of 99.19%, precision of 99.28%, and an F1-score of 0.9772. These results confirm the effectiveness of the proposed SOS-CNN hybrid approach in improving classification accuracy.