Every method we have discussed so far is able to detect patterns in past data. These patterns will be useful for predicting the future, assuming that the future looks mostly like the past. As the cliché goes, “Correlation is not causation.” But correlation might be all that you need. If people who rent the movie Frozen also tend to like the movie Sing, then you can reliably use this information for future recommendations. Both preferences are likely caused by a third unseen signal—that these people have small children who like cartoon musicals—but that doesn’t lessen the usefulness of the discovered pattern.
However, when you are ...