NETWORK INTRUSION DETECTION SYSTEMS USING CONVOLUTIONAL NEURAL NETWORK

Abstract

Our work aimed to develop the detection of attacks using intrusion detection systems (IDSs). We achieved this by developing a model capable of accurately classifying different types of attacks, merging convolutional neural network (CNN) and auto-encoder capabilities, and applying transfer learning (TL) on the target model. This integration proved to be highly effective in enhancing the performance of our IDS. By leveraging the power of CNNs, we achieved remarkable results, with a validation accuracy of 99.97% and an impressively low validation loss of 0.0022. Furthermore, we explored the potential of TL to further enhance our model’s capabilities. Initially, when we applied TL without fine-tuning, the validation accuracy of our target model was 60.41%, accompanied by a validation loss of 0.6600. However, by implementing fine-tuning techniques, we observed a significant improvement in performance. The proposed model, after undergoing fine-tuning, exhibited a remarkable validation accuracy of 99.85%, with a further reduction in validation loss to 0.0068. Through continuous training on various attack datasets, we ensured that our model becomes increasingly adept at recognizing and classifying diverse attacks. Our approach allows for the efficient detection of attacks while reducing the training time required for the model. As a result, we have succeeded in creating a robust and adaptable intrusion detection system capable of identifying different types of attacks with high precision.