Predicting bankruptcy is the focus of our research, which is one of the important aspects of research, as due to bankruptcy both company’s goodwill and shareholders’ benefits are affected. In order to predict bankruptcy, reliable models are required. The focus of this paper is based on different deep learning models. However, developing deep learning models for forecasting bankruptcy is one of the challenging tasks as most of the datasets are imbalanced in nature. So we first try to balance the dataset. US Bankruptcy Prediction Data set (1971-2017) is taken here, which is very imbalanced in nature. To balance the dataset both undersampling and oversampling and one hybrid method are used. In this research, a comparison is made among three different types of models like CNN, LSTM and ANN by applying all balancing techniques on each classifying model, for the prediction of bankruptcy. Here we get that the ANN model gives better results than the other two whatever balancing technique may be used and among the balancing techniques oversampling is far better than undersampling, but here the hybrid sampling method outperformed all in each classification model as compared to the others.