Network attacks detection and classification via deep learning algorithms

This thesis investigates the use of deep learning techniques to automatically detect and classify network-based attacks. Traditional intrusion detection systems often rely on hand-crafted rules or shallow classifiers, which struggle to adapt to evolving threat profiles. We propose end-to-end architectures—leveraging CNNs for spatial pattern recognition in network traffic features and RNNs (e.g., LSTM) for temporal sequence modeling—to identify anomalies and categorize attacks (e.g., DoS, Probe, U2R, R2L). Models will be trained and evaluated on standard datasets (NSL-KDD, CICIDS2017), with preprocessing steps including feature normalization and dimensionality reduction. Expected outcomes include high detection accuracy, low false-positive rates, and insights into feature importance for different attack classes.