Introduction to Machine Learning

What is machine learning?

Imagine teaching a child to recognize cats. You probably wouldn't give a perfect rule like "cats have exactly this ear angle and this tail length." You'd show many examples. Machine learning (ML) works similarly: instead of hand-writing every rule, we give a computer data, and it learns patterns that help it make decisions or predictions.

At its core, ML is about building a model: a system that maps inputs to outputs.

• Input: an email message

• Output: spam or not spam

• Input: house size, location, number of bedrooms

• Output: predicted price

A simple way to think about ML is:

Experience + patterns = better predictions

Unlike traditional programming, where rules are explicitly coded, ML lets the computer infer useful rules from examples.

How learning from data works

Suppose we want to predict apartment rent. We collect examples with known answers:

• 500 sq ft → $1200
• 700 sq ft → $1600
• 900 sq ft → $2100

A model might learn a rough relationship like:

predicted rent = 2 × square_feet + 200

Worked example:

For an 800 sq ft apartment,

predicted rent = 2 × 800 + 200 = 1800

The model is rarely perfect, so we measure error:

error = actual - predicted

If the real rent is $1850, then:

error = 1850 - 1800 = 50

Learning means adjusting the model to reduce this error across many examples. In practice, ML systems repeat this process thousands or millions of times.

square_feet = 800
predicted_rent = 2 * square_feet + 200
print(predicted_rent)  # 1800

Two common kinds of machine learning

One major category is supervised learning. Here, the model trains on examples with correct answers, called labels.

Examples:

• Medical scan → tumor or no tumor
• Past sales data → next month's demand
• Transaction details → fraud or not fraud

Another category is unsupervised learning, where the data has no labels. The model looks for structure on its own.

Examples:

• Grouping shoppers into similar customer segments
• Finding unusual network activity
• Organizing news articles by topic

A quick test: if you know the target answer during training, it's probably supervised. If you're asking the computer to discover patterns without answer keys, it's probably unsupervised. Both are useful, but they solve different kinds of problems.

Where you already see machine learning

ML is woven into daily life, often invisibly. When Gmail filters spam, Spotify recommends songs, or Google Maps estimates travel time, a model is using patterns from past data.

Common applications include:

• Recommendation systems: YouTube, Netflix, Amazon
• Image recognition: face unlock, medical imaging, self-driving perception
• Language tools: translation, autocorrect, chatbots
• Forecasting: weather, sales, energy demand
• Risk detection: fraud alerts, credit scoring, cybersecurity

But ML is not automatically trustworthy. If training data is biased, outdated, or too small, the model can make poor decisions. A face recognition system trained mostly on one demographic, for example, may perform worse on others.

The big idea: ML is powerful when patterns exist, data is relevant, and results are checked carefully.