2
$\begingroup$

I'm reviewing Chapter 6 from An Introduction to Statistical Learning. I'm having trouble understanding PCA and the provided example.

Can someone explain how the first principal component direction can yield observation projections with the largest possible variance, but also minimize the sum of the squared perpendicular distances between each point and the line?

In other words: how does the green line both maximize variance and fit the data as closely as possible? Seems to me that projections onto the blue line would have higher variance.

For reference, here's the excerpt from the book and the example figures:

The first principal component direction of the data is that along which the observations vary the most. For instance, consider Figure 6.14, which shows population size (pop) in thens of thousands of people, and ad spending for a particular company (ad) in thousands of dollars, for 100 cities.

The green solid line represents the first principal component direction of the data.

If we projected the 100 observations onto this line (as shown in the left hand panel of Figure 6.15), then the resulting projected observations would have the largest possible variance; projecting the observations onto any other line would yield projected observations with lower variance.

There is another interpretation for PCA: The first principal component vector defines the line that is as close as possible to the data. In Figure 6.14, the first principal component line minimizes the sum of the squared perpendicular distances between each point and the line. These distances are plotted as dashed line segments in the left-hand panel of 6.15, in which the crosses represent the projection of each point onto the first principal component line.



6.14enter image description here

6.15enter image description here

$\endgroup$
4
$\begingroup$

Your reference already gives the answer.

The first principal component direction of the data is that along which the observations vary the most

This is referring to the projections of the data onto that line, sort of the variance explained by that line.

I think you might be interpreting this something like:

The first principal component direction is the dimension on which the residuals vary the most

which is really the opposite statement: the residuals are the deviations from that line, which is the variance of everything that is not the first principal component.

The first principal component is the one that captures as much of the variance as possible along that dimension.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.