What machine learning is, in plain words
Machine learning teaches a computer to make predictions or decisions by finding patterns in data, instead of following rules someone wrote by hand.
Think of teaching a student. You show them examples, they work out the pattern, and later they apply it to a situation they have never seen. The machine does the same thing, only the examples are rows in a file.
Why the data matters more than the model
Data is the experience the machine learns from. Well-labelled, representative data produces a useful model. Messy or one-sided data produces a confident, wrong one — and no amount of tuning fixes that.
Most datasets are a table: each row is one example, each column is a feature the model can learn from. Here is the shape of a small one.
| id | age | salary | education | purchased |
|---|---|---|---|---|
| 1 | 29 | 52,000 | Bachelor | Yes |
| 2 | 45 | 85,000 | Master | No |
| 3 | 37 | 61,000 | Diploma | Yes |
The last column is what you are trying to predict. The rest are what the model gets to look at.
The seven types
-
Supervised learning
The machine learns from examples that include both the input and the correct answer. Learning with a teacher. Used for classification (is this email spam?) and regression (what will this house sell for?).
-
Unsupervised learning
Only inputs, no answers. The machine has to find the structure itself — grouping similar customers, or simplifying data while keeping what matters.
-
Semi-supervised learning
A small amount of labelled data plus a large amount of unlabelled data. Useful when labelling is expensive but raw data is easy to collect.
-
Reinforcement learning
The machine acts in an environment and gets rewards or penalties back. Like training a dog with treats. Common in robotics, games and control systems.
-
Self-supervised learning
The model creates its own labels from the data — hide part of the input, predict the hidden part. This is how most large language and vision models are pretrained.
-
Deep learning
Multi-layered neural networks that learn complex patterns. Not a separate category so much as a technique you can apply inside any of the above. It wins when data and compute are plentiful.
-
Transfer learning
Take a model trained on one task and reuse it as the starting point for a related one. It saves time and data because the generic features are already learned.
Which one to pick
Start from what you have. Labelled examples means supervised. No labels means unsupervised or self-supervised. A problem involving a sequence of decisions means reinforcement. Anything involving images, audio or text usually means deep learning or transfer learning.
Good machine learning is roughly 80% data and 20% modelling. Collect, clean and understand your data first.