Machine Learning Defenses: Protecting Financial Markets from Ai-Driven Attacks

Abstract

This research focuses on exploring different machine learning defence strategies in protecting the financial market from attacks by AI. Based on the assessment of the main defence mechanisms, including adversarial training, anomaly detection, and model robustness, this work identifies strategies that help to minimize the threats linked to malicious AI manipulation. The results presented prove that adversarial training enhances model robustness at the cost of accuracy and that autoencoders with a suitable architecture are highly effective in detecting anomalous behaviour at the cost of high time consumption. It also identifies how the three factors; accuracy, latency and efficiency are all interrelated and trade-off against each other especially when developing models for real-time decision making in the complex financial world. In addition, updating the model regularly is described as crucial to sustaining strong protective measures across the duration of threats. In summary, this work offers several contributions for enhancing the understanding of how machine learning can be used to better prevent and combat future threats in financial markets, and presents a clear path for how this protective research can be advanced further.