Prerequisites

• ML Introduction with scikit-learn
• Linear Regression with Python

Classification is a supervised learning task. Its goal is to predict the class to which the instance belongs based on a set of parameters (features). You need to give many labeled examples of data (called training set) for the computer to learn before it can predict the class of a new instance.

The difference between classification and regression is that regression predicts a continuous numerical value, for example, a price. It can be any real (only positive for a price) number.

In contrast, classification predicts a categorical value, for example, the type of a sweet. There is a finite set of values, and the model tries to classify each instance into one of these categories.

Based on the formulation of a problem, there are several types of classification:

• Binary classification: in binary classification, a target is one of two possible outcomes. For example, email: spam/not spam, sweet: cookie/not cookie;

• Multi-class classification: in multi-class classification, there are three or more possible outcomes for a target. For example, email: spam/important/ad/other, sweet: cookie/marshmallow/candy;

• Multi-label classification: in multi-label classification, each instance can belong to multiple classes at the same time. For example, a movie can be classified as both action and comedy, or an email can be marked as both important and work-related.

For most ML models, you need to encode the target to a number. For binary classification, outcomes are usually encoded as 0/1 (e.g., 1 - cookie, 0 - not a cookie). For a multi-class classification, outcomes are usually encoded as 0, 1, 2, ... (e.g., 0 - candy, 1 - cookie, 2 - marshmallow).

Many different models can perform classification. Some examples include:

• k-Nearest Neighbors;
• Logistic Regression;
• Decision Tree;
• Random Forest.

Luckily, they are all implemented in the scikit-learn library and are easy to use.