Deep Learning-Based Cybersecurity Framework for Proactive Intrusion Detection on the CSE-CIC-IDS2018 Dataset

Abstract

Network infrastructure has become more complex and the amount of cybersecurity challenges has grown with the introduction of cloud computing, the Internet of Things (IoT) and next-generation communication technologies. The known attack and known signatures are the reasons why traditional IDSs fail to detect sophisticated and new attacks; they are signature-based and programmed by hand. To overcome these drawbacks, this work presents a deep learning-based cybersecurity framework for proactively detecting the intrusion in CSE-CIC-IDS2018 benchmark dataset. The proposed system combines spatial and temporal features of network traffic using a hybrid Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) network architecture with a systematic data preprocessing pipeline. In the preprocessing stage, data quality and learning efficiency are enhanced by data cleaning, handling missing data, removing duplicate data, encoding labels, normalizing features, and splitting the data into training and testing sets. The CNN part is used to automatically learn discriminative spatial features, and the LSTM was used to learn the sequential traffic dependencies to see the relation between the sequential sequence and attack detection capability was improved. Experiments were carried out on Python, TensorFlow, Keras and Google Colab with GPU support. The proposed framework was evaluated according to the accuracy, precision, recall, F1-score, and ROC-AUC and the results were found to be 98.64%, 98.41%, 98.18%, 98.29%, and 99.12%, respectively, which are considered to be excellent classification results and strong discrimination ability. Moreover, the framework demonstrated good computational efficiency with an average inference time of 2.8ms per network flow and a moderate use of GPU memory. Results of comparative analysis with the latest deep learning-based intrusion detection methods also validate the competitiveness and applicability of the proposed framework. The results highlight the efficiency, reliability, and effectiveness of the hybrid CNN-LSTM framework in proactively detecting network intrusions in contemporary cybersecurity settings.