DEEP LEARNING BASED INTRUSION DETECTION SYSTEM

Abstract

The rapid growth of computer networks and IoT devices has significantly increased the risk of cyberattacks, emphasizing the importance of strong network security to ensure data integrity, availability, and confidentiality. Intrusion Detection Systems (IDS) play a vital role in identifying unauthorized access and malicious activities. However, traditional IDS methods, such as signature-based and anomaly-based approaches, often struggle to adapt to the continuously evolving threat landscape. This study introduces Deep-IDS, a hybrid model that combines Convolutional Neural Networks (CNN) for spatial feature extraction and Long Short-Term Memory (LSTM) networks for capturing temporal dependencies. Techniques such as batch normalization and dropout are applied to enhance generalization and minimize overfitting. Deep-IDS is trained and evaluated on the NSL-KDD and CICIDS2017 datasets, achieving high detection accuracy, low false alarm rates, and strong overall performance. Furthermore, Deep-IDS is deployed as a Flask-based web application, providing real-time monitoring and interactive user engagement, offering an effective and scalable solution to defend against emerging cyber threats in IoT environments.

Author

Dr. R. Barona, Anto Nismi C, Jebisha A, Jenitta J