Tip:
Highlight text to annotate it
X
Welcome back to the last installment of the ten new features in STATISTICA 10 in 10
billion series buddy mistaken for thompson and so they'll be talking to
you about profit chart and raptor
receiver operating characteristic errors
these clubs are used in particular analytics in data mining project to
evaluate models and determine how to move forward with the models created
before we look at these two plots i want to set up the example that were using
i'm using their credit scoring example data set to build predictive models
the variable of interest credit rating
is at classification variable with to outcome good or bad
so our customer state into one of two categories
they had a good credit rating meaning they paid back their credit in full and
on time
where they had a bad credit rating meaning they missed payments or default
my objective is that create a predictive
model class trying new customers as good or bad credit rating using several
variables
including
their payment of previous credit
and how long they were employed by their current employer
the predictive models have already been built using data mining tools
now i'm at the point of evaluating the models and applying them
with the credit scoring data
will select the data mining menu
and she was rapid deployment
both the models
in here i can select multiple models that for simplicity and the senate
select one model
the rain forest
predictive model
and now select their profits are and rockers scam
anne pursuant to appropriate values for the creation of this profit chart
so first i changed the category variable
to get it
the profit charge should be in terms of good credit risk customers because this
is where we make our money
for the cost per observation weekend at the cost of a missed classification
that means when we contend creditor customer
who will default
for each customer it is
value is going to be different forgiving you can add rich dot you so let's say on
average
when a customer default the concept of danny
one thousand dollars
correct classifications earned a profit of four hundred dollars
now select profits are to make the plot
we actually get to pieces of output we get the spreadsheet
of outlet for the profits are
and then the plot itself
this plot says expected profits
using the rainforest model
verses the basic model
backs access shows the customers in percentile so using the reinforced model
to customers are ranked by the model's predictions have a good credit rating
for each customer review was correctly classified profit increase by four
hundred dollars
for each missed vacation profits decreased by thousands
so that's why she says is that
when using the random forest model
we expect high profits when extending credit to customers predicted to be a
good credit risk customer
back on the rapid deployment dialogue
let's make the raptor
and again we get two pieces about that we get the spreadsheet
and in the plot
receiver operating characteristic stickers rockers
where virtually developed during world war two
radar is used to detect airplanes that sometimes a bird which show that
on radar and look like an airplane
so the rocker visually showed accuracy but
plotting the true positives by the false-positive
the true positive would be when and airplane was correctly identified
a false-positive is when
hubbard was falsely identified as an airplane
now the curves are used to evaluate model performance as well
for classification type miles
the plot shows how well the model does ac correctly classifying it category
in our example the catagories get credit risk customers
the more area between the base rate line
and the line for the predictive model
the better
the models predicted accurately for at that category
immediately reacted for given in the title as well.
If you'd like more information about StatSoft and its products including
STATISTICA, visit our website at statsoft.com or call
918-749-1119
Outside of the United States and Canada statsoft.com has links to our
international offices that can also be of assistance to you
and thank you for watching the series.