adjusted

I'm high.

I'm captain.

Right? And we're talking about multiple in your regression.

So what is it?

Some class we covered simple linear regression.

And that is basically where this one independent variable to

one dependent variable.

Simple, you know, But a multiple linear Russian is different.

Where is there's more than one independent variable and all

of that in relation to the just one dependent variable.

So a good example that would help me understand this

concept was, um, the cost of the house.

So ah, house.

The cost of it is affected by a lot of

different things, such as like where it's located, the area

of it's a good area versus bad area of the

bedrooms, five bedroom versus one bedroom.

I'm school zoning that's in.

It's good good school.

Go to good schools on breast, the bat schools.

All those a fact.

The price of the house, which is the dependent variable

at the factors of the independent variable.

So we just gives us a better understanding of how

they came to price of the house.

Okay, we've got assumptions.

So some of these are some of the assumptions that

need to be done.

So the main one is multi Colin narrative.

Sorry, but basically, that means that the end, uh, between

the independent variables, there shouldn't be Ah ah, high correlation,

like with them within them.

Um and so this, um, is one of the assumptions,

and the Medalla would be independent Visit jewel, and then

residuals must be normally distributed siem mean of zeros in

her direction.

The maze it's on then.

Furthermore, the relationship between, um each pretty predictor variable an

outcome variable is layer.

So until these air done good to go, um, next

about our square so r squared in multiple.

In your regression, there is a few problems with it,

um, one of them being that every time you add

a, uh, every time you add a variable looking independent

bearable the R squared of increases like even if it's

by chance, it increases every single time.

And guess causes a really high are square of ice

r squared.

And, uh, that is misleading and just wrong because, um,

it's going up based on the independent variables, not based

on the accuracy.

Um, Also another, a problem is over fitting.

So over fitting a model is where, um, it randomly

um, the model, like the statistic model randomly starts toe.

Um, describe the data, random error data and, um, the

random their error data, then the relationship between the variables.

So this problem occurs when the model is too complex

and also leads to ah, uh, misleading high R squared

value. And it's just wrong.

Um, so let's go to the next slide.

Something we can do we can use to get a

more accurate R squared.

Value is the adjusted our square.

Um, so they adjust.

Our square is a modified version of the R squared

that it has been adjusted to, uh, get the number

of predictors in the model.

So I've put this little picture here really useful in

explaining this.

So if we look, uh, with the bear, like each

time it increases like 12345 the R squared value just

keeps increasing, but on the r squared adjusted, you can

see how it hits a peak at number three on,

and then it just goes back down.

So it's a real.

It's like an accurate value of the R squared.

Um, so that's why we should use the Are Split

and Justin why it's superior to the arse word.

Um, okay, you met covered pretty much this.

Um so basically, it's not that Oh, something else.

It's not better.

Um, like, if we use if we if we use

our if you use the r squared that I'll you

look, if we look at the r squared value instead

of looking at the r sport adjusted, it doesn't mean

since its higher doesn't mean it's better in any way.

It's misleading.

And it's wrong because, um, it's going up each time

we are another independent like another.

Um predictor Pretty.

Yeah, Another predictor.

It's just misleading.

Inbal. Okay, moving up.

Um, now time for the example of good example.

So good example would be right now you don't see

the charts.

We don't understand what I'm saying, but this is me

explaining. Turn for him.

It is weight and horsepower, and it's in relation to

the mileage and that's response.

So the, uh, predictors are the horsepower and wait and

the response is mileage.

So now let me show you so in, Uh uh,

in this type of model in a multiple linear regression

model, you can't make it to D of mass me

three year because it is two different exes, and so

it's kind of it comes out to be a three

D model, and it's a little harder for humans to

visualize that it's three d model, um, in comparison to

a Judy model.

The grocery.

Um, so this right here is a really good example

of it.

Um, you can see the mpg right here.

And then you could see the weight in the horsepower

at the bottom, and then the regression line, all that.

So this is too good money.

Now, have you got okay, So what does it matter?

Why does it a good model matter?

Why do we even why do we even me to

use multiple linear regression model?

And why can't we always just use the simple linear

one? Because it's, you know, easier to do.

So basically, in real life, there is never sorry Maxim.

Tonight I keep looking at the, um there is never

in real life is never just like a 1 to

1 factor.

There's so many different, um, things that happen.

Like like, for example, the house, like there's just so

many different variables that happened.

That result to this one outcome.

So if we have a better way of capturing a

more accurate, um, response like a more out accurate outcome

with all these other factors, then we're already, like, you

know, one step ahead.

So it allows us, you know, to planning and, um,

control. We can put, um, yet planning control and then,

um, weaken prediction and forecasting.

We can predict things based on all these other

adjusted

Audio TranscriptAuto-generated