Exists a very famous dataset called MNIST digits classification. This dataset contains images of handwritten digits from 0 to 9. Each instance is a 784 values in range 0 to 255 that represents the pixel intensity at a given location(every image is 28x28 size).
But here I'm gonna try a different approach. Instead of use "784 features" (28x28), I'm gonna apply a Dimension Reduction and use only 2 features that will be calculated Intensity and Symmetry of the images.
Other different aproach is that I'm gonna use only linear classifiers to classify the digits. I'm gonna use Linear Regression, Logistic Regression and Perceptron all algorithms made by me. And classify the digits 0, 1, 4 and 5.
Disclaimer: The data was previous manipulated to keep a "default behavior" that will be seen in the Notebook.