Enhancing Cloud Security with AI-Driven Anomaly Detection for Zero-Day Threats

Advancements in communication technology facilitate the transmission of diverse data across network systems in cloud environments. Simultaneously, as communication technology progresses, the attack surface widens, heightening concerns about network security. Consequently, extensive research focuses on network intrusion detection systems (NIDS) to address potential threats. Among NIDS technologies, there's growing interest in AI-based anomaly detection, leveraging Deep Learning (DL) and Machine Learning (ML) techniques to enhance the development of Intrusion Detection System (IDS). Industries increasingly demand resilient IDSs capable of detecting zero-day attacks. However, data imbalance poses a persistent challenge, hindering accurate detection of zero-day attacks by AI models. This study introduces a novel AI-based NIDS addressing data imbalance and enhancing performance. An autoencoder with both basic and deep architectural implementation is proposed for zero-day attack detection, aiming for high recall and minimizing false-negatives. To validate the effectiveness of our model, we compare its performance against that of standard methods. Evaluation on renowned IDS datasets—CICIDS2017 and UNSW-NB15—shows promising results, with autoencoders effectively identifying intricate zero-day attacks. This proposed model achieves a detection accuracy of over 98% on both datasets.