Machine Learning: 2 Books in 1: Machine Learning for Beginners, Machine Learning Mathematics. An Introduction Guide to Understand Data Science Through the Business Application

Spam Detection
A common example of a relatively simple machine learning tool is spam
detection. If we are using supervised learning, defining the variables that
are relevant, then the model will have certain characteristics to look for in
email messages received. The model might look for certain keywords or
phrases in order to detect if an email is spam or not. Words like ‘buy’ or
‘save’ might let your inbox know when you are receiving spam email. The
problem with this method is there are many cases in which those words
might not necessarily mean spam. There might be other keywords or
combinations of words that we would overlook.
This is where reinforcement learning comes in. There are so many
characteristics that may be indicative of spam email, some of them we may
not even be able to explain. Reinforcement learning will allow to model to
find these patterns on its own, without explicit guidance. Instead, we tell the
model when it has identified spam correctly. Sometimes we find an email
message in our inbox that the model didn’t classify as spam, so we move it
to our spam folder manually. Now the model knows that this message is

---

spam, and this piece of data is added to the model to improve the prediction
next time. So over time, the machine improves as it gets more relevant data.
This type of machine learning is known as classification. Our output falls
into discrete categories. In statistics, discrete variables are variables which
can be identified in only a finite number of categories. An example of a
discrete variable would be the number of cars sold by a car dealership in a
week. It’s discrete because the car dealership can’t sell half a car. The
variable must be a whole number.

---