Scholarlink

In this modern digital environment vast volume of data are being generated and communicated over the network to make the complex task easier but it also has raised the cybersecurity threats extensively. Thus intrusion detection solutions for network environments are used to continuously track the data flowing in the network and raise alarm when any malicious activity is detected. Integration of machine earning and deep learning have made these network intrusion detection systems (NIDS) more capable and an essential component in the domain of cybersecurity. This study investigates a unified model built by fusing ‘Convolutional Neural Network and Bi-directional Long-Short Term Memory Network’ (CNN + BiLSTM) based NIDS. It has been tested against the CSE-CIC-IDS2018 dataset which consists of recent and realistic threats instances. This hybrid model is efficient for spatial feature extraction as well as temporal sequence learning. To evaluate its effectiveness, the model is compared against two hybrid Deep learning based approaches namely CNN combined with stacked LSTM and another is the combination of Gated Recurrent Unit and Multilayer Perceptron (GRU + MLP). This study demonstrated the high accuracy of 99.11% and improved detection of less frequent cybersecurity threat (web application threats) by CNN + BiLSTM in comparison to the other two models due to the advantage of convolutional and time-series components of the model.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing the Network Intrusion Detection Using CNN–BiLSTM: A Comparative Study with CNN–Stacked LSTM and MLP–GRU Architectures

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ajay Kumar Kushwaha,
  • Pawan Kumar Tiwari

摘要

In this modern digital environment vast volume of data are being generated and communicated over the network to make the complex task easier but it also has raised the cybersecurity threats extensively. Thus intrusion detection solutions for network environments are used to continuously track the data flowing in the network and raise alarm when any malicious activity is detected. Integration of machine earning and deep learning have made these network intrusion detection systems (NIDS) more capable and an essential component in the domain of cybersecurity. This study investigates a unified model built by fusing ‘Convolutional Neural Network and Bi-directional Long-Short Term Memory Network’ (CNN + BiLSTM) based NIDS. It has been tested against the CSE-CIC-IDS2018 dataset which consists of recent and realistic threats instances. This hybrid model is efficient for spatial feature extraction as well as temporal sequence learning. To evaluate its effectiveness, the model is compared against two hybrid Deep learning based approaches namely CNN combined with stacked LSTM and another is the combination of Gated Recurrent Unit and Multilayer Perceptron (GRU + MLP). This study demonstrated the high accuracy of 99.11% and improved detection of less frequent cybersecurity threat (web application threats) by CNN + BiLSTM in comparison to the other two models due to the advantage of convolutional and time-series components of the model.