Given a dataset containing information on miles per gallon, number of cylinders, and such of various cars, can we predict what the value for miles per gallon would be if we only had the other values available?

In this case, our outcome is mpg and we are using the other variables of cyl (Cylinders), hp (Horsepower), gear (number of gears), and others to build a model that can then be applied against a dataset where the values for mpg are marked as MISSING. The model reads the information in these columns in the first five rows of the data and, based on that information, predicts what the value for mpg would be in the other rows, as shown in the following image:

The reason this is considered supervised is that in the course of building our machine learning model, we provided the model with information on what the outcome was. Other examples include:

• Recognizing letters and numbers: In such cases, the input to the model are the images, say of letters and numbers, and the outcome is the alpha-numeric value shown on the image. Once the model is built, it can then be used against pictures to recognize and predict what numbers are shown in the picture. A simple example, but very powerful. Imagine if you were given 100,000 images of houses with house numbers. The manual way of identifying the house numbers would be to go through each image individually and write down the numbers. A machine learning model allows us to completely automate the entire operation. Instead of having to manually go through individual images, you could simply run the model against the images and get the results in a very short amount of time.
• Self-driving autonomous cars: The input to the algorithms are images where the objects in the image have been identified, for example, person, street sign, car, trees, shops, and other elements. The algorithm learns to recognize and differentiate among different elements once a sufficient number of images have been shown and thereafter given an unlabeled image, that is, an image where the objects have not been identified is able to recognize them individually. To be fair, this is a highly simplified explanation of a very complex topic, but the overall principle is the same.

MNIST Dataset used for number recognition: