Although we discussed Bayes' theorem, we would be doing a great disservice to you if we did not discuss Naive Bayes. It's everywhere, and for good reasons. It almost always works well (hence the name, Naive), and you will most certainly be exposed to it during your machine learning career. It is a simplistic technique based upon a premise that the value of any one feature is completely independent from the value of any other. For example, an orange is round, the color is orange, the skin is not smooth, and it's 10-20 cm in diameter. A Naive Bayes classifier would then consider each feature described previously to contribute independently that this is an orange versus an apple, lemon, and so on, even if there is some data relationship amongst its features.

As mentioned, Naïve Bayes is surprisingly efficient in resolving complex situations. Although there are scenarios where it can certainly be outperformed, it can be a great first-try algorithm to apply to your problem. We only need a very small amount of training data compared to many other models.