An Empirical Study Using Microsoft Azure Auto Machine Learning to Detect Zero-Day Attacks
Joy Buuri, Samah Mansour, Mostafa El-Said, Xinli Wang · 2024
Evolving zero-day attacks in network security pose a significant challenge, as they remain unknown to the network systems and security professionals, rendering them exceptionally hazardous. Despite advancements in cybersecurity technologies, existing approaches to combating zero-day attacks often face limitations regarding predictive accuracy, scalability, and adaptability to evolving threats. Traditional models may struggle to keep pace with the dynamic nature of zero-day attacks, leaving organizations vulnerable to exploitation. Additionally, traditional machine-learning approaches may rely on static models and manual tuning. In this paper, we harness the power of Microsoft Azure automated machine learning (AutoML)'s dynamic learning capabilities and adaptive responses for seamless model deployment and continuous adaptation to emerging threats. Experimental analysis was conducted to evaluate the accuracy and effectiveness of Azure AutoML in detecting zero-day anomalies. We measure the model's performance through multiple datasets and rigorous testing and validation processes, including precision, recall, F-measure, and overall AutoML model accuracy.