Microarrays are sophisticated datasets that have a substantial number of features and a limited number of samples. This can lead to a class imbalance problem. Balancing class distribution is an important research domain. In this paper, a novel variational auto encoder model is proposed for class imbalance mitigation. The proposed model employs an adversarial loss function to generate simulated data. The proposed model is evaluated using the autism gene expression dataset. The obtained experimental results prove that the data generated from the proposed model have obtained noteworthy results compared with the original dataset by employing three different classifiers.