The diagnosis of rare muscular disorder such as Facioscapulohumeral Muscular Dystrophy (FSHD) using gene expression data through machine learning and deep learning presents a major challenge due to limited data availability. This leads to overfitting and poor generalization capability of the model. To address this problem our study presents a Wavelet transformation - generative adversarial network (WT-GAN) based data augmentation model (MicroNMDNet) for micro-array dataset. We used publicly available FSHD dataset (GEO accession GSE 39398) which contains gene expression data of 50 samples. Out of which 26 were FSHD affected samples and 24 were healthy samples. As the micro-array dataset has high dimensions(33,297) we applied wavelet transform to decompose each gene expression signal into approximation (cA) and detail (cD) coefficients. These components retain important aspects and pattern from original dataset. MicroNMDNet-GAN a WT-GAN architecture is used where generator tries to generate realistic approximation coefficients (cA) from random noise and discriminator tries to differentiate between real cA and generated cA. Detail coefficient (cD) is sampled from training data in a class balanced manner (25 FSHD + 25 normal) and combined with generated cA. After applying inverse wavelet transformation 50 new synthetic gene expression samples are created which are further merged with original dataset. The augmented data is used to train classifiers to differentiate between FSHD and healthy samples. The model was evaluated using metrics such as Accuracy, F1-score, ROC-AUC and confusion metrics. Results showed that performance of model improved when data augmentation is applied on the dataset using WT-GAN. In summary, our study demonstrates that data augmentation with WT-GAN can yield promising results which offers a promising approach for bio-informatics and precision medicine tasks.