The increasing adoption of IoT devices in smart homes demands stronger security due to their vulnerability to cyber-attacks. Traditional anomaly detection methods often struggle with class imbalance in IoT datasets, where certain attack types are rare. This paper addresses this challenge by proposing a novel methodology that improves anomaly detection in smart home IoT networks. We use class-specific Generative Adversarial Networks to generate artificial data, enriching the representation of minority classes and creating a more balanced training set. Each anomaly class is mapped to a unique GAN, enhancing dataset diversity. Our approach significantly outperforms traditional methods, showing improvements in detection metrics like accuracy, precision, recall, F1-score, and ROC-AUC. This highlights the efficiency of our method in overcoming class imbalance in IoT environments.