[MUSIC] Now that we've taken a deep
dive into a linear classifier, namely logistic regression, let's jump
into a different kind of classifier. One called a decision tree,
which is extremely useful in practice, especially when it combines
with something called boosting, which you're going to see
in an upcoming module. In this module, we're going to jump in
to another really interesting real world example, which falls more into
the financial data modelling part. It turns out that decision
trees are extremely useful and used a lot in the finance industry, so
we're going to jump into something in that domain to give you a little bit more
experience to that kind of data, in particular when you look at
evaluating loan applications. So lets say, I want to buy a house. I'm excited,
I want to buy this particular house. To buy it, I don't have all the money,
I need to take a loan from the bank. So the bank is going to look at some
properties of my history, like my credit, what has it been like in the past? How much money I make? How long my loan is of which time I'm
willing to spend before I pay it off? And other personal information about me,
like my gender, age and so on. And he's going to take
that information and try to make a prediction as to whether
loaning me money is a risky thing or not. So, let me give you a little bit
more details on the kinds of things that people measure when they try
to make loan application decisions. So typically, you look at credit history, which looks at all the other loans I've
taken in the past and my credit cards and all that and
have I paid those off on time. Then we look at my income,
how much money do I make today? And we'll also look at what's called the
term of the loan, which may be 3 years, 5 years, 15 years, 30 years, which is the period of time
I'm going to take to pay back that loan. So, let's take a three-year loan
means I pay back within three years. And finally, a look at information about
the particular individual like my age, whether married and so on. So, given this information, I want to make a prediction as to whether
loaning me money is a risky thing or not. So, a loan application system
might look something like this. You get as input all the loan information
that fill up a bunch of forms. It goes through a system if we
have machine learning system or maybe minor created one. Hopefully, machine loan system, which is going to make a prediction
whether this loan is safe. It's okay to make that loan or
whether it's risky. So, some applications is going to be
classified as safe while others will come in and be classified as risky. And the bank may be unevenly making
loans to risky applications, but often to save funds and we can view this loan
application as a classification problem. I'm given this input information
from my form and let's called the x, all information about me and my loan. I pushed it through the classifier
modelm which going to make a decision. Let say y plus 1, y height equal plus 1,
if it's a safe loan and y height equals minus 1
if it's a risky loan. In this module, we're going to use what
is called decision tress classifier. It might look a little bit like this,
you start the application and look at some particular feature
of the particular input. Let's say, what has my credit been like? If my credit has been excellent, I just make the loan without looking
at any other information about me. But if my credit has been only fair,
I look at the term of the loan. If the term is short,
maybe it's too risky. But if the loan is long,
maybe I'll take enough time to pay it off. Now if my credit is poor,
I don't stop there. I look further where my income is even
if my credit didn't pass being poor by making a bunch of money,
maybe I'm willing to make a long-term loan to somebody with poor credit
that makes a bunch of money. But if you don't make that much money and
your credit is poor, then you're out of luck. No loans for you. [MUSIC]