As the data is images, therefore I have used Convolutional Neural Network. There are two convolutional layers followed by one pooling layer and a dropout layer, then there is another convolutional layer followed by pooling layer and a dropout layer.
After flattening two hidden layers have been used with 1024 and 256 neural units respectively.
Finally for output there are ten classes and softmax activation has been used for best accuracy.
I have saved the results in a pickle file and have also saved the model.