Thoughtful Machine Learning with Python

Chapter 4. Naive Bayesian Classification

Remember how email was several years ago? You probably recall your inbox being full of spam messages ranging from Nigerian princes wanting to pawn off money to pharmaceutical advertisements. It became such a major issue that we spent most of our time filtering spam.

Nowadays we spend a lot less time filtering spam than we used to, thanks to Gmail and tools like SpamAssassin. Using a method called a Naive Bayesian Classifier, such tools have been able to mitigate the influx of spam to our inboxes. This chapter will explore that topic as well as:

• Bayes’ theorem

• What a Naive Bayesian Classifier is and why it’s called “naive”

• How to build a spam filter using a Naive Bayesian Classifier

As noted in Table 2-2, a Naive Bayes Classifier is a supervised and probabilistic learning method. It does well with data in which the inputs are independent from one another. It also prefers problems where the probability of any attribute is greater than zero.

Using Bayes’ Theorem to Find Fraudulent Orders

Imagine you’re running an online store and lately you’ve been overrun with fraudulent orders. You estimate that about 10% of all orders coming in are fraudulent. In other words, in 10% of orders, people are stealing from you. Now of course you want to mitigate this by reducing the fraudulent orders, but you are facing a conundrum.

Every month you receive at least 1,000 orders, and if you were to check every single one, you’d spend more money fighting ...