Mankind has always hoped to do better. Ensemble learning[12] is one of the techniques that help drastically improve results of machine learning algorithms. But these classifiers need a lot of data in order to perform really well. But even in today’s data driven world, it is not that easy to get hold of that magnitude of data. This is where the concept of synthetic data[2] comes into play. Synthetic data is not real, but boasts many characteristics of real data, and the latter can sometimes become indistinguishable from the former. This data can be put to good use, such as training our ensemble classifiers, so that we can get even better results. That is what this paper aims to illustrate. I’m using random-forest ensemble classifier for classifying MNIST handwritten digits dataset, and the Generative Adversarial Network (GAN) for generating the said synthetic data.