As it's implied in its name, Machine Learning (ML) is the science of building machines (algorithms) that can learn from data. In other words, this class of algorithms generates certain outcomes (predictions) based on the relations they infer from the training data—not from the hardcoded, predetermined rules. Usually, ML is described as having two main branches—supervised and unsupervised ML.

Unsupervised models attempt to find structure in the data itself, without any given supervision or target to focus on. The usual task is to find clusters of similar records (for example, users) to understand the underlying latent logic (for example, using target audiences and the corresponding use cases for the service). 

Supervised learning is all about training the model by feeding it pairs of independent features and the correct values of the target variable of interest as a training set. For example, supervised ML is used to detect fraudulent activity, given a user's actions or to get an estimate of a certain value (for example, the price of a house)—all by inferring the result from the training dataset, which includes the target variable.

Many models use complex math and require huge computation power, but that is not always the case—some of them are very simple to use and easy to comprehend. Most importantly, ML runs solely on mathematics; although it might be incredibly useful and it can empower and enable, it can never replace common sense and critical thinking. 

Let's now go around and see how different supervised and unsupervised models can be trained and used to analyze our WWII dataset.