<p>The continuous evolution of attacker methods to evade security systems makes fraud detection in cybersecurity an essential problem. Standard rule-based fraud detection techniques fail to detect sophisticated fraudulent activities. In this study, an AI-based system is presented, which simulates user behavior in order to detect fraud in machine learning. The model uses historical logs of user activities which include IP addresses amounts failed to transact and the time of the user to identify the presence of possible fraudulent activity. Isolation Forest algorithm can be used to identify abnormal habits of the user, separating legitimate and fraudulent actions. The model proves efficient detection of fraud cases and low false positive having been tested on artificial and actual dataset samples. The article explains why deep learning models such as LSTMs and autoencoders can be used to improve anomaly detection. The suggested system allows detecting fraud in real-time that reinforces cybersecurity protection by forecasting threats prior to their development.</p>