H1B Visa Analysis using GANs

The H1B visa is a non-immigrant visa that enables foreign employees to enter the country and work there for a set amount of time. The purpose of the study is to analyze the 5 years of data on H1B Visa petitions from 2011-2016 and record the findings. Machine Learning algorithms will be employed to predict if the H1B Application Case status. The project also applies the functionality of Generative Adversarial Networks (GANs) to augment the training data. Tabular GANs are applied to the H1B Visa data and after augmentation, to test it on regular GANs, it is converted to images. These are fed to the generator network of a regular GAN model tested on MNIST data. The results show that GANs decrease the accuracy and increase the randomness of the data: Logistic Regression (before GAN=87.18%, after=73.8%), Random Forest (before=80.5%, after=70%), Gradient Boosting Ensemble (before=87.87%, after=71.3%), KNN (before=85.8%, after =71.4%), ANN (Error score Before GAN = 0.656, After GAN=1.033). Images formed by GANs to match the MNIST data showed satisfactory results after training for 5000+ epochs.