Before getting our hands dirty, we should discuss strengths and weakness related to tree-based models. To begin with strengths, trees are often inexpensive to train and understand (things can get complicated for large trees). Just as training tree models don't usually require much of computational power, understanding how the model is rolling often won't require more than figuring out a simple dendrogram.

As they can be understood with little effort, these models are called glass-box. Through a simple visualization, almost anyone can see how such a model is engaging in decisions. Despite their simplicity, tree models are very flexible, meaning they are capable of fitting linear and non-linear relations.

This leads us to a downside. Flexibility comes from the ways that a tree can be specified or misspecified. Misspecifying will only cause doom and overfitting. On the other hand, this disadvantage is turned into an advantage when it comes to random forests (more details on those later). If we have no intention of trusting random forests, we can also use some techniques such as bagging, boosting, or pruning. 

Tree-based models are a kind of supervised learning and can be used to solve many queries. In order to get practical, we could try ranking bank customers as good and bad payers. Else, we could try to predict prices of commodity prices. Instead, we will be trying to predict voting intentions for the Chilean plebiscite of 1988.