Javad Zabihi’s research while affiliated with University of Maryland, Baltimore County and other places

Anomaly detection has been one of the most interesting research areas in the field of cybersecurity. Supervised anomaly detection systems have not been practical and effective enough in real-world scenarios. As a result, different unsupervised anomaly detection pipelines have gained more attention due to their effectiveness. Autoencoders are one of the most powerful unsupervised approaches which can be used to analyze complex and large-scale datasets. This study proposes a method called DeepNet, which investigates the potential of adopting an unsupervised deep learning approach by proposing an autoencoder architecture to detect network intrusion. An autoencoder approach is implemented on network-based data while taking different architectures into account. We provide a comprehensive comparison of the effectiveness of different schemes. Due to the unique methodology of autoencoders, specific methods have been suggested to evaluate the performance of proposed models. The results of this study can be used as a foundation to build a robust anomaly detection system with an unsupervised approach.