00:01
Once again, welcome to new problems.
00:05
This time we're dealing with matrices and these matrices are found in statistical modeling of what we can call statistical models.
00:23
So you're dealing with statistical models.
00:26
In this instance, when we talk about classification, classification, matrix, it's the same reference as a confusion matrix.
00:42
So a classification matrix can be synonymous with a confusion matrix.
00:49
And this is found in machine learning.
00:53
So when you look at machine learning, this is what can happen.
00:59
Sometimes the classification matrix or confusion matrix is known.
01:03
As the error matrix and it visualizes the performance of an algorithm and this type of algorithm is a supervised learning algorithm so supervised learning algorithm so this is a classification matrix that you're saying right here and it's a standard tool it's a standard tool for evaluation of statistical models so it's a standard two for evaluation for statistical models with the chart you have a chart that compares actual actual to predicted values for each specified, for each specified predicted state.
02:31
So if you're going to specify a certain predicted state.
02:37
We have a new problem and in this particular problem we're looking at insurance records and we want to develop a model.
02:52
Want to develop a model for predicting fraudulent claims.
03:05
So in an insurance we want to develop a model for predicting fraudulent claims.
03:15
If you look at historical database, then 1 % are judged to be fraudulent.
03:26
Fraudulent.
03:30
I got to be fraudulent.
03:31
So this is when you're looking at the claims.
03:34
So we have a sample database.
03:40
We have a sample database.
03:43
And this is used to develop.
03:46
It's used to develop a model.
03:53
And the process of the sampling is introduced.
04:05
Used to provide balanced samples.
04:17
To provide balanced samples.
04:20
And this happens because of low response response rates.
04:31
So you're going to have low response rates that we want to do over sampling to control for this.
04:38
The total number of records the total number of records and is 800 correct classification is 310 frauds and 279 frauds as the correct classification involves 90 frauds, that's the mis -classification involving 90 frauds.
05:28
Incorrect, incorrect, involves 130 records.
05:39
So we have certain questions and requirements create a classification matrix, matrix for the sample, greater classification matrix for the sample, determine the adjusted misclassification rate, meaning you're adjusting for the sampling, so determine the adjusted misclassification rate, determine the adjustment rate, determine the percentage of new records, of new records, deemed classified as fraudulent.
06:51
So these are the questions we're dealing with in the problem, the one is that in part a, we're looking at the classification matrix, we already have those numbers.
07:04
In the classification matrix cases are sorted into categories cases sorted into categories checking if the predicted value matched the actual value.
07:49
All the cases are then counted and totals placed in a matrix.
08:05
So 1 % real frauds are mentioned, meaning 8 out of 800 were real frauds, 8 out of 800 were real frauds...