The MNIST ( Modified National Institute of Standards and Technology) database is a large collection of handwritten digits as monochrome images. The digits have been size-normalized and centered in a fixed-size image. The goal of this experiment is to find a set of hyperparameters that result in an accurate model and excellent model performance using GridSearchCV from Scikit-learn as a tunning technique. Randomly I selected 3 optimizers (Adam, RMSprop, and SGD) as a starting point to develop the MLP models; later I tuned the 3 different models with their own hyperparameters and chose the model that predicted with higher accuracy the digits. I created my own database with 20 samples and I preproced the images trying to simulate the original set to predict digits with the selected model. In the end, the model showed high accuracy on the test set but the accuracy decreased with the custom dataset. The problem with MNIST is that the dataset is "too perfect"; in rea...