Decision Trees
Explore how decision trees are used in classification tasks, especially in finance for predicting loan defaults. Understand the greedy approach to building trees, selecting the best features to split data, decision stumps, and stopping criteria to manage tree size and accuracy.
We'll cover the following...
Decision trees are commonly used in financial domains due to their ability to solve and explain prediction problems. They are also used as a module in other algorithms to solve more complex algorithms like bagging and boosting.
To understand the decision trees, let’s look at the problem of a car loan defaulter prediction. When we want to get a loan from the bank, they ask some questions and derive answers from our past credit history. Questions may relate to:
- Monthly income
- Personal information
- Previous loans
- Current properties
Let’s say a person has a monthly income of $10K. Their age is 37. They have not taken any previous loans and do not own any property. A decision tree with the past data (from other customers) can be created like this:
This is a very basic example. In the real world, a lot of other and complex data points are considered.
So, according to the data, this person $(Income< ...