Machine learning means that instead of a programmer writing a rule for every case, we show the computer many examples so it can discover the pattern itself. The result of this process is a "model" that can make predictions or decisions on new data.
A simple example#
Say you want to filter spam emails. Writing manual rules for every spammer trick is almost impossible. With machine learning, you give the algorithm thousands of labeled emails (spam or legitimate). The model learns which words and patterns look suspicious and classifies new emails on its own.
The steps#
- Collect and clean the data.
- Choose a type of algorithm.
- Train the model on part of the data.
- Evaluate the model on data it has not seen before.
- Use it in the real world and monitor it continuously.
Three main families#
- Supervised learning: the data has labels.
- Unsupervised learning: the data has no labels and the model finds groups itself.
- Reinforcement learning: the model learns through rewards and penalties.
Where you meet it#
Product recommendations in an online store, fraud detection in bank transactions, sales forecasting and content filtering are all machine learning.
Common mistake#
A model is never better than its data. If the data is scarce, incomplete or biased, even the most accurate algorithm gives weak results.