A type of machine learning where an algorithm learns from labeled examples to produce a model that can make predictions on new data and therefore learn without supervision.

How It Works

1. Give the algorithm a dataset
2. Each example in the dataset contains data + label (inputs + outputs)
3. The algorithm looks at many examples to find the relationship between the data and labels.
4. Once trained the model can take new data (input) and predict the label (output).

Simple Example

A model to identify spam is provided with 1M examples of spam email and 1M examples of legitimate email. Each example is labelled spam or legit. An algorithm can learn what patterns make up spam and what patterns are not. This will produce a model that can identify spam emails.

Key Techniques

• Regression - predicting a continuous value (e.g. number, time, cost)
• Classification - Predicting a label or category (e.g. spam or not spam)

Algorithms

• Decision Trees - Splitting data in to branches based on yes/no questions
• Support Vector Machines (SVM) - Finding the best boundary to separate categories
• Neural Networks - Layers of nodes similar to neurons that understand complex data
• K-Nearest Neighbors (K-NN) - Classifying new data by looking at the closest example