Practical Predictive Analytics

Unsupervised learning models

Unsupervised problems are more exploratory in nature. In an unsupervised learning context, one doesn't specify a target variable, or even have any idea what something should be. A data scientist is usually given a set of attributes, and then asked to derive some relationships, or discover some new attributes which are not obvious from looking at the data.

As an example, during an exploratory customer analysis, you could look at different attributes of customers, such as age, gender, and sales history, which could then lead you to create new variables (which hadn't existed before), which would describe each customer as best, good, or average. You could then choose an appropriate algorithm to suit this problem which ...