AI Specialization Program

WEEK 2  Lesson 5 

Title:

Feature Engineering Basics

Theme

Turning Raw Data into Machine Intelligence

Why This Lesson Exists

Most beginners believe better models create better AI.

In reality:

Good features beat complex models.

A simple model with strong features often outperforms a complex neural network trained on poor data.

Feature engineering is where domain understanding meets machine learning.

This lesson teaches students how to transform raw data into signals models can actually learn from.

Lesson Objective

By the end of this lesson, students will:

✔ Understand why features matter more than algorithms

✔ Apply scaling and normalization correctly

✔ Encode categorical data for ML models

✔ Extract basic features from text data

✔ Develop intuition about feature importance

Lesson Structure.

1. What is a Feature?

A feature is any measurable property used by a model.

Example — Loan Prediction Dataset

Raw Data           |     Feature Used by Model

  Age                 |     Numeric feature

  City                 |     Encoded category

  Salary              |     Scaled numeric

Transaction history    Aggregated feature

Models do not understand meaning — only numbers.

Your job:  Convert reality into numbers intelligently.

2. Why Features Matter More Than Models

Demonstration Concept.

Same dataset:

Model A: Logistic Regression + strong features → 85% accuracy

Model B: Neural Network + poor features → 65% accuracy

Lesson:

AI success = Data Representation Quality.

Key Principle:

Performance ≈ Data Quality × Feature Quality × Model Choice

Not the other way around.

3. Scaling & Normalization

Many algorithms assume comparable numeric ranges.

Problem Example:

Feature        |     Value Range

  Income      |  10,000–2,000,000

    Age         |        18–60

Income dominates learning unfairly.

Standard Scaling (Z-score)

Centers data around zero.

Python

from sklearn.preprocessing import StandardScaler

scaler = StandardScaler()

df[['income','age']] = scaler.fit_transform(df[['income','age']])

Min-Max Normalization

Rescales between 0 and 1.

Python

from sklearn.preprocessing import MinMaxScaler

scaler = MinMaxScaler()

df[['income']] = scaler.fit_transform(df[['income']])

When to Scale

✓ KNN

✓ Neural networks

✓ Gradient descent models

Usually unnecessary for:

❌ Tree-based models (Random Forest, XGBoost)

4. Encoding Categorical Variables

Machines cannot read text categories directly.

One-Hot Encoding

Python

pd.get_dummies(df['city'])

Result:

Lagos  |   Abuja    |   Accra

    1             0  |      |   0

Best for:

Low-cardinality categories.

Label Encoding

Python

from sklearn.preprocessing import LabelEncoder

le = LabelEncoder()

df['gender'] = le.fit_transform(df['gender'])

Use carefully — may introduce false ordering.

Practical African Context

Common categorical data:

Language

Region

Payment type

Device type

Network provider

Students learn to encode responsibly.

5. Basic Text Feature Extraction

Text must also become numbers.

Bag of Words

Python

from sklearn.feature_extraction.text import CountVectorizer

vectorizer = CountVectorizer()

X = vectorizer.fit_transform(df['message'])

Counts word frequency.

TF-IDF (Better Baseline)

Python

from sklearn.feature_extraction.text import TfidfVectorizer

tfidf = TfidfVectorizer()

X = tfidf.fit_transform(df['message'])

Highlights meaningful words.

Example Use Cases

SMS fraud detection

Customer feedback analysis

Social media sentiment

Support ticket classification.

6. Feature Importance Intuition

Not all features help equally.

Students learn to ask:

Which variables actually influence predictions?

Which features add noise?

Example (Tree Model)

Python

model.feature_importances_

Visualization helps explain models.

Why This Matters

Feature importance enables:

✔ Explainable AI

✔ Bias detection

✔ Better data collection decisions

7. Feature Engineering Mindset

Students adopt this thinking loop:

Observe data

↓

Hypothesize useful signal

↓

Create feature

↓

Test performance

↓

Iterate

AI development becomes experimental science, not guesswork.

Mini Practice Exercise

Students will:

Load dataset.

Encode categorical columns.

Scale numeric features.

Create one derived feature:

Example: income_per_age.

Train simple model and compare results.

Common Beginner Mistakes (Important Section)

❌ Scaling before train/test split (data leakage)

❌ Encoding categories inconsistently

❌ Creating too many useless features

❌ Ignoring domain knowledge

Real-World Insight

In industry:

Data scientists spend 60–80% of time on feature engineering and data preparation — not modeling.

This lesson reveals the reality of AI work.

Lesson Outcome

By the end of Lesson 5, students can:

✔ Transform raw datasets into ML-ready features

✔ Apply scaling and normalization correctly

✔ Encode categorical variables safely

✔ Extract numerical features from text

✔ Understand why feature design drives model performance

Next: 

 Week 2 — Lesson 6: Model Lifecycle & Production Thinking

—the integration lesson that turns students from learners into AI system thinkers

Powered by Soft AI Africa | Training the Next Generation of AI Leaders in Africa.