Machine Learning Basics in Simple Words

Rules vs learning — the core difference

A traditional program follows rules a human wrote: “if the email contains these words, mark it spam”. The problem: spammers change words daily. Machine Learning (ML) flips the job — you show the computer thousands of examples of spam and not-spam, and it discovers the patterns itself. Rules written by data, not by hand.

The vocabulary that unlocks everything

• Data — the examples: photos, emails, sales records, anything measurable
• Features — the measurable details the model looks at: word counts in an email, brightness patterns in a photo
• Labels — the answers for training examples: “spam” or “not spam”, “cat” or “dog”
• Model — the trained result: a file full of adjusted numbers that can now judge new examples
• Training — the learning phase, where the model guesses, checks its errors, and adjusts — millions of times

The three styles of learning

• Supervised learning — trained with labelled answers; used for spam filters, price prediction, disease screening support
• Unsupervised learning — no labels; the model finds groups by itself, like sorting customers into natural segments
• Reinforcement learning — learning by trial, error and reward; famous in game-playing AI and robotics

A worked example — predicting study success

Imagine data from past students: hours studied per week, attendance percentage, and the final result (pass/fail) as the label. A supervised model trains on this, finds that low attendance plus low hours strongly predict failure, and can then flag current students who match the risky pattern — early enough for a teacher to help. Same recipe, endless dishes: swap the data and the label, and you get a loan-risk model, a crop-health model, or a recommendation engine.

What ML cannot do

Models learn from the data they are given — biased or incomplete data produces biased, fragile models. A model also cannot explain itself like a human or handle situations completely unlike its training. Knowing these limits is part of using ML professionally.

Practice task (45 minutes)

Design an ML system on paper: pick a problem you care about, list the data you would collect, the features, the label, and one risk if the data is biased. This design thinking — before any code — is exactly how real ML projects begin.