Data Analysis From Scratch With Python: Step By Step Guide

11. Classification
Spam or not spam? This is one of the most popular uses and examples of
Classification. Just like Regression, Classification is also under Supervised
Learning. Our model learns from labelled data (“with supervision”). Then, our
system applies that learning to new dataset.
For example, we have a dataset with different email messages and each one was
labelled either Spam or Not Spam. Our model might then find patterns or
commonalities among email messages that are marked Spam. When performing
a prediction, our model might try to find those patterns and commonalities in
new email messages.
There are different approaches in doing successful Classification. Let’s discuss a
few of them:
Logistic Regression
In many Classification tasks, the goal is to determine whether it’s 0 or 1 using
two independent variables. For example, given that the Age and Estimated
Salary determine an outcome such as when the person purchased or not, how can
we successfully create a model that shows their relationships and use that for
prediction?
This sounds confusing which is why it’s always best to look at an example: 
Here our two variables are Age and Estimated Salary. Each data point is then
classified either as 0 (didn’t buy) or 1 (bought). There’s a line that separates the
two (with color legends for easy visualization). This approach (Logistic

---

Regression) is based on probability (e.g. the probability of a data point if it’s a 0
or 1).
As with Regression in the previous chapter wherein there’s this so-called black
box, the behind the scenes of Logistic Regression for Classification can seem
complex. Good news is its implementation is straightforward especially when
we use Python and scikit-learn: Here’s a peek of the dataset first
(‘Social_Network_Ads.csv’):