Deep Learning-Based AI Attack Detection: A Real-World Cybersecurity Dataset Approach

Abstract

Advanced AI-driven attack detection systems have been advanced in reply to the complexity of cyber threats in the era of Artificial Intelligence (AI). Nevertheless, strong and adaptable cybersecurity solutions are required since adversarial AI attacks are constantly changing. To increase the accuracy of threat detection, this paper presents a framework for AI attack detection that is based on deep learning (DL). The proposed research makes use of two real-world cybersecurity datasets, such as UNSW-NB15 and CICIDS2017. This study works enhanced than other Machine Learning (ML)-based Intrusion Detection Systems (IDS) because it employs a combination of Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM), and Transformer architectures to find complex cyberattacks. An evaluation with conventional security models reveals higher detection rates, increased efficiency and lower false positive rates. Aside from that, testing shows how well DL models withstand hostile AI threats like poisoning and evasion. The results guarantee better protection against ever-changing cyber threats by outlining a course of action for AI-driven cybersecurity defenses.