Anomaly detection plays an increasingly vital role in modern industrial systems, where automation and sensor integration are essential. These systems rely on sensor data and threshold-based algorithms to detect unusual patterns that may indicate potential faults or failures. The integration of machine learning, particularly supervised learning approaches, has significantly improved detection accuracy by training models on labeled datasets that distinguish between normal and abnormal conditions. However, a major challenge remains: the scarcity of abnormal data. Since such events are rare, collecting sufficient samples for training machine learning models is often time-consuming and costly. To address this, we propose a generative AI-based method to synthesize training data that includes rare abnormal scenarios. This synthetic dataset helps build models that generalize better and avoid overfitting. The trained models are then validated on a real-world, unseen dataset, and performance metrics such as false positives and false negatives are analyzed. Consequently, our approach indicates the potential of using generative AI agents to synthesize the rare event to assist the better training process of several machine learning models as Multi-Layer Perceptron can reach up to 97.8% accuracy on actual data while being trained on the synthetic data of Llama.