Fine-Tuning CNN-BiGRU for Intrusion Detection with SMOTE Optimization Using Optuna

Abstract

Network security faces a significant challenge in developing effective models for intrusion detection within network systems. Network Intrusion Detection Systems (NIDS) are vital for protecting network traffic and preempting potential attacks by identifying signatures and rule violations. This research aims to enhance intrusion detection using Deep learning techniques, particularly by employing the NSLKDD dataset to train and evaluate a hybrid CNN-BiGRU algorithm. Additionally, we utilize the Synthetic Minority Over-sampling Technique (SMOTE) to address imbalanced data and Optuna for fine-tuning the algorithm's parameters specific to NIDS requirements. The hybrid CNN-BiGRU algorithm is trained and evaluated on the NSLKDD dataset, incorporating SMOTE to tackle imbalanced data issues. Optuna is utilized to optimize the algorithm's parameters for improved performance in intrusion detection. Experimental results demonstrate that our approach surpasses classical intrusion detection models. Achieving an accuracy rate of 98,83 % on NSLKDD, the proposed model excels in identifying minority attacks while maintaining a low false positive rate. The findings affirm the efficacy of our proposed approach in network intrusion detection, showcasing its ability to effectively discern patterns in network traffic and outperform traditional models