Summary  
This chapter introduces data preprocessing techniques including exploring dataset features, addressing missing values, encoding categorical variables into numerical form, and recognizing the need for feature scaling.

General domain of usage  
Machine learning

Begin preprocessing by exploring the dataset. Throughout this course, the **penguin dataset** will be used, with the task of predicting the species of a penguin.


There are three possible options, often referred to as **classes** in machine learning:

The features are: `'island'`, `'culmen_depth_mm'`, `'flipper_length_mm'`, `'body_mass_g'`, and `'sex'`.

The dataset is stored in the `penguins.csv` file. It can be loaded from a link with the `pd.read_csv()` function to examine its contents:


import pandas as pd

df = pd.read_csv('https://codefinity-content-media.s3.eu-west-1.amazonaws.com/a65bbc96-309e-4df9-a790-a1eb8c815a1c/penguins.csv')

print(df.head(10))

This dataset presents several issues that need to be addressed:

* Missing data;
* Categorical variables;
* Different feature scales.


## Missing Data

Most ML algorithms cannot process missing values directly, so these must be addressed before training. Missing values can either be **removed** or **imputed** (replaced with substitute values).

In `pandas`, empty cells are represented as `NaN`. Many ML models will raise an error if the dataset contains even a single `NaN`.


## Categorical Data

The dataset includes categorical variables, which machine learning models are unable to process directly.


Categorical data must be **encoded** into numerical form.


## Different Scales

`'culmen_depth_mm'` values range from 13.1 to 21.5, while `'body_mass_g'` values range from 2700 to 6300. Because of that, some ML models may consider the `'body_mass_g'` feature **much more important** than `'culmen_depth_mm'`.

**Scaling** solves this problem. It will be covered in later chapters.

Machine learning drives modern technological innovation across all industries. Embark on a comprehensive introduction to predictive modeling by mastering foundational algorithmic concepts utilizing Scikit-Learn. Participants will construct robust classifiers, evaluate performance metrics, ultimately culminating in the development of a complete predictive project.

Getting Familiar with Dataset

Missing Data

Categorical Data

Different Scales