As cyber threats increase in sophistication and stealth, traditional Intrusion Detection Systems (IDSs) typically experience difficulties identifying emerging types of attacks (e.g., zero-day attacks) because they rely on known signatures and lack interpretability. This paper presents the development of SENTRY-AI, an explainable, multi-modal anomaly detection framework that integrates deep learning and computer vision to improve cybersecurity defenses in cyberspace. SENTRY-AI architecture employs a Variational Autoencoder (VAE) to conduct unsupervised anomaly detection on tabular network features, in addition to a Convolutional Neural Network (CNN) to analyze time-series traffic data transformed into a Gramian Angular Field (GAF). An analysis is performed with a late-fusion approach utilizing outputs from the two models to evaluate the efficacy and robustness of the hybrid approach.
