In this project we performed a supervised prediction task on the Modified-MNIST dataset, which is a more challenging variant of the original MNIST. Our goal was to output the digit in the image with the highest numeric value. We developed convolutional neural network architectures for our model and also used famous image classification models: VGG16, MobileNet, and ResNet-50. Also in some models we used methods such as data augmentation to further improve the accuracy. We obtained the best performance with VGG16 network with a validation accuracy of 99.03% and an accuracy of 98.933% on Kaggle test set.