Paper Contents
Abstract
Anomaly-based detection systems identify potential threats by monitoring network behavior and detecting deviations from expected patterns. Although many models have been proposed, there is a gap in research when it comes to evaluating these models across a range of publicly available datasets. As cyber threats evolve quickly, it is essential to consistently update and benchmark intrusion detection datasets. Traditional methods for multi-class intrusion detection, such as deep neural networks, often fail to recognize spatial relationships and long-term dependencies in traffic data. This project introduces a new deep learning approach that addresses these challenges and aims to build a dependable system for detecting cyberattacks. Our framework includes three core strategies, the first being an autoencoder integrated with various optimization techniques. Experimental results demonstrate the effectiveness of this hybrid model in identifying contemporary threats. The CICIDS2017 dataset was used to test the model, which successfully classified multiple types of attacks. The model showed strong performance, particularly with the Adamax optimizer, in terms of accuracy, detection rate, and minimizing false alarms. Comparative analysis confirms that our solution outperforms other machine learning and deep learning models in these key areas.
Copyright
Copyright © 2025 Yuva Bhargav Mandapati. This is an open access article distributed under the Creative Commons Attribution License.
