Adversarial Attacks on Machine Learning Models in Cybersecurity

This research paper aims to convey the robustness and importance of machine learning and offer insights on defense mechanisms against data poisoning. In the past few years, machine learning models have shown an increase in the deployment of them in real-world applications. Our research shows that despite the increase use of machine learning, it still remains vulnerable to adversarial attacks. The research evaluates the effects of adversarial attacks on a convolutional neural network (CNN) that was trained on a MNIST dataset. we have successfully conducted 3 experiments Fast gradient sign method (FGSM), Label flipping attack, and confusion matrix comparison.