Association rules

The last unsupervised approach we're considering is based on the discovery of association rules and it's extremely important in the field of data mining. A common scenario is represented by a collection of commercial transactions made up of a subset of products. The goal is to find out the most important associations between products (for example, the probability of buying P_i and P_j is 70%). Specific algorithms can efficiently mine a whole database, highlighting all the relationships that can be taken into account for strategic and logistic purposes. For example, an online store can employ this method to promote all those items that are frequently bought together with other ones. Moreover, a predictive approach allows simplifying the provisioning processes by suggesting all those products that are very likely to be sold out, thanks to an increase in the sales of other items.

At this point, it's helpful to introduce the reader to a practical example of unsupervised learning. No particular prerequisites are needed, but it's preferable to have a basic knowledge of probability theory.