Landowners Hate Him!!! This Data Scientist Discovered One Weird Trick to Improve Trump’s Wall!

Now, we all know that Donald J. Trump is a man of many great ideas. But, alas, even a man as perfect as Trump occasionally has his flaws. After much hard work, I was able to uncover that one of his iconic proposals was obtained without, shockingly enough, any input from Support Vector Machines(SVM)!

This, of course, is an egregious error:

An Introduction to SVMs

In models like SVM, you’re given a dataset with some features about a variable, plus their appropriate responses (eg red and blue).

And you’re trying to see if you get more data with features, you can correctly classify the new points as red or blue.

In the simplest SVM, your training data will have discrete clusters. So eg, you’ll have a clump of red points on one end, and a clump of blue points on the other. Assume that you have added data with features, but not the responses, how do you identify whether the new points will likely be red or blue?

One thing you can do is to just draw a line between the two clumps.

Diagrams from Introduction to Statistical Learning by Gareth James

And assume that one side is red, and the other side is blue.

But if the clumps are fully separable , there are infinitely many borders that you can draw that will work.

So how do you pick the right line?

Well, the standard way to do this is called the “maximal margin classifier”, which draws a line in the middle that is as far away from both the nearest red and nearest blue points as possible.

In more advanced models, the reds and blue will not be as discretely separated, so you have to have some margin of error.

So eg, you might have some blue points in the red side of the border, or red points in the blue side of the border.

The point however is to try to find the right border that maximizes the probability that blue on the blue side and red is on the red side, even for points where you don’t currently have data on whether they’re red or blue .

How This Applies to Trump’s Wall

Now, Donald J. Trump wants to build a wall, for the express purpose of separating the Americans from the Mexicans. But he arbitrarily picks national borders selected in the pre-computer and pre-Machine Learning era…and egregiously (I consider this Trump’s greatest flaw) forgets to implement Best Practices from Machine Learning

Thus, I would like to advise candidate Trump in his new wall building.

The Great Wall of America should use Support Vector Machines to identify the right boundaries which maximizes Mexicans on one side of the wall and American citizens on the other.

According to a preliminary analysis, the Binary Classification Wall should be built across half of Texas, Arizona, New Mexico and California.

Questions for Further Study

Now, some people may quite absurdly claim that restricting immigration with a wall costing $25 billion is economically and morally unsound. But those complainers are at worst total haters and losers, and at best, rendered completely irrelevant by the latest advances in machine learning. An yuge, optimal, SVM-guided wall design, will of course be sufficient for even the strongest naysayers.

After all, who cares if you’re optimizing for the wrong thing, as long as you’re doing it efficiently?

A more interesting question is the sophistication of our SVM model. A simple model will draw a linear or quasi-linear line, as demonstrated above. But sometimes a linear model is insufficient… an improved version of Trump’s wall, using kernels, will have boundaries within cities, for example blocking ethnic ghettos in the US or American tourism spots in Mexico. This is a question that deserves further research, and my think tank will be happy to accept additional donations.

Finally, some of the naysayers are landowners who might be slightly perturbed by a wall cutting across the middle of their property, or preventing them from accessing the nearest grocery store, hospital or Donald Trump Rally. Fortunately, this is a problem that already exists:

Thus, I will claim that the difference between my proposed improved boundaries and Trump’s current wall is only one of degree, and not of kind.

So Donald Trump, if you’re reading this, I have only one last thing to say: Mr.Trump, please hire me!!! Because without Machine Learning, how else can we #MakeAmericaGreatAgain?