Last Updated on June 4, 2026 by Rajeev Bagra

When starting machine learning on Kaggle, you may see code like:

import pandas as pd
from sklearn.metrics import mean_absolute_error
from sklearn.model_selection import train_test_split
from sklearn.tree import DecisionTreeRegressor

At first glance, this looks like a collection of mysterious imports. In reality, each line represents an important concept in the machine learning workflow.

Let’s understand them one by one.

---

What Is a Library?

A library is a collection of pre-written code that developers can reuse.

Instead of writing thousands of lines of code yourself, you import libraries created by experts.

Example:

import pandas as pd

This imports the Pandas library.

Without Pandas, reading and manipulating datasets would be much more difficult.

---

What Is Pandas?

Pandas is the most popular Python library for working with data.

It helps you:

• Read CSV files
• Filter rows
• Select columns
• Calculate statistics
• Clean data

Example:

import pandas as pd

The pd is simply a nickname.

Now instead of writing:

pandas.read_csv()

you can write:

pd.read_csv()

---

What Is a Dataset?

Machine learning learns from data.

A dataset is simply a collection of information.

Example:

House Size	Bedrooms	Price
1500	3	250000
1800	4	320000
2200	5	450000

This table is a dataset.

---

What Is a CSV File?

CSV stands for:

Comma-Separated Values

Example:

size,bedrooms,price
1500,3,250000
1800,4,320000
2200,5,450000

The Kaggle code loads data from:

iowa_file_path = '../input/home-data-for-ml-course/train.csv'

This file contains housing data used for prediction.

---

What Is Reading Data?

The dataset is loaded into memory using:

home_data = pd.read_csv(iowa_file_path)

This tells Pandas:

Open the CSV file and convert it into a table that Python can work with.

---

What Is a DataFrame?

A DataFrame is Pandas’ table structure.

Example:

home_data

might display:

LotArea	YearBuilt	SalePrice
8450	2003	208500
9600	1976	181500

Think of a DataFrame as a spreadsheet inside Python.

---

Features and Target Variables

Machine learning learns relationships.

Suppose:

Size	Bedrooms	Price
1500	3	250000

The inputs:

Size
Bedrooms

are called features.

The output:

Price

is called the target variable.

---

Features (X)

Features are information used to make predictions.

Example:

features = ['LotArea',
            'YearBuilt',
            '1stFlrSF']

---

Target (y)

The value we want to predict.

Example:

y = home_data.SalePrice

Here the model tries to predict:

SalePrice

---

What Is Machine Learning?

Traditional programming:

Rules + Data = Answers

Machine learning:

Data + Answers = Rules

The algorithm discovers patterns automatically.

---

What Is a Model?

A model is the learned pattern.

Example:

Larger house → Higher price
Newer house → Higher price

The model learns such relationships from data.

---

What Is Scikit-Learn?

Scikit-Learn is Python’s most popular machine learning library.

Import example:

from sklearn.tree import DecisionTreeRegressor

It provides:

• Decision Trees
• Linear Regression
• Random Forests
• Clustering
• Metrics

and much more.

---

What Is a Decision Tree?

The screenshot imports:

from sklearn.tree import DecisionTreeRegressor

A Decision Tree makes decisions by asking questions.

Example:

Is house size > 2000?
          |
        Yes
          |
    Price > 300000

        No
          |
    Price < 300000

The model repeatedly splits data into smaller groups.

---

Why Is It Called a Regressor?

There are two major ML tasks.

Regression

Predict a number.

Examples:

• House price
• Temperature
• Salary

Example output:

$350,000

---

Classification

Predict a category.

Examples:

• Spam or not spam
• Cat or dog
• Pass or fail

Example output:

Spam

Since house prices are numerical values, Kaggle uses:

DecisionTreeRegressor

---

What Is Training a Model?

Training means allowing the model to learn from data.

Example:

model.fit(X, y)

The model studies:

Features → SalePrice

relationships.

---

What Is Prediction?

After training:

model.predict(X)

The model guesses outputs.

Example:

Actual Price: 250000
Predicted Price: 245000

---

Why Split Data?

The screenshot imports:

from sklearn.model_selection import train_test_split

Why?

Because testing on the same data used for learning is misleading.

A student who memorizes answers is not necessarily knowledgeable.

Similarly, a model must be tested on unseen data.

---

What Is Training Data?

Training data teaches the model.

Example:

80% of data

---

What Is Validation Data?

Validation data tests the model.

Example:

20% of data

The model has never seen this data before.

---

train_test_split()

Example:

train_X, val_X, train_y, val_y = train_test_split(
    X,
    y,
    random_state=1
)

This automatically creates:

• Training features
• Validation features
• Training targets
• Validation targets

---

What Is random_state?

Example:

random_state=1

This makes results reproducible.

Without it:

Split #1 → Different rows
Split #2 → Different rows

With it:

Always same split

Useful for debugging and learning.

---

What Are Metrics?

The screenshot imports:

from sklearn.metrics import mean_absolute_error

A metric measures model performance.

Think:

Exam Score for ML Models

---

Mean Absolute Error (MAE)

MAE measures average prediction error.

Example:

Actual	Predicted
100	105
200	190
300	310

Errors:

5
10
10

Average:

8.33

MAE = 8.33

Lower is better.

---

Underfitting

A model that is too simple.

Example:

All houses cost $200,000

Clearly unrealistic.

The model misses important patterns.

---

Overfitting

A model that memorizes training data.

Example:

House #1 → exactly $253,871
House #2 → exactly $311,422

Works perfectly on training data.

Fails on new data.

---

The Goal: Generalization

A good model should:

• Learn patterns
• Avoid memorization
• Predict accurately on unseen data

This ability is called:

Generalization

and it is the ultimate goal of machine learning.

---

The Entire Kaggle Workflow in One Diagram

CSV Dataset
      │
      ▼
Pandas DataFrame
      │
      ▼
Select Features (X)
Select Target (y)
      │
      ▼
train_test_split()
      │
      ├── Training Data
      └── Validation Data
      │
      ▼
DecisionTreeRegressor
      │
      ▼
Train Model
      │
      ▼
Predict Prices
      │
      ▼
mean_absolute_error()
      │
      ▼
Evaluate Performance
      │
      ▼
Tune Model
      │
      ▼
Avoid Underfitting & Overfitting

This workflow is the foundation not only of Kaggle’s introductory course but also of most real-world machine learning projects.