Machine learning and age

the mindlab

I’ve been using Mathematica more for work, and finally getting better at it. It’s only been 20 years…

Anyway, Mathematica has some machine learning capabilities built in, so I started playing around with them. Maybe something would be useful for work? (Probably not.) Anyway, a few of the tools have pre-trained datasets to play with, so you can just plug stuff into them.

One is the “NameAge” dataset, which predicts the most likely age for a given name. Naturally, I put some names in. I then plotted up the difference between the predicted age and the person’s actual age. In my chart, negative numbers represent how much younger the person is predicted to be compared to their actual age, and positive numbers are older.

Buff does the worst, but with her given name, she is predicted to be much younger. Up next is Joe, Mike, Pete, Tom, and myself; according to this machine learning dataset, we should be getting ready for retirement, clocking in at more than 15 years older than our actual age.

7 thoughts on “Machine learning and age”

  1. I’m not sure that I am interpreting correctly. Are you saying that I should be 7 years younger? But Ruth is such an OLD name, which seems like I should be 10 years older, whereas you guys’ names are younger names. Or should all of you be Tyler, Adam, Garrett, etc. in order to seem young?

  3. Weird… Krissy and Al are the closest. And yeah, what is up with the tie between Tom and I? You should have spelled his name Thom.

  4. Ma Gray, re: your age prediction, guessing the age based on the name doesn’t take into account the departed. Take a look at this chart: http://nametrends.net/name.php?name=ruth I would say that most Ruths at the height of 1915 aren’t still around (even worse: http://nametrends.net/name.php?name=ethel ).

    On top of this, rate isn’t total number. Ruth dropped in popularity from 0.07524% in 1940 to 0.03645% in 1952, which looks like less than half of babies were being named that, but when you put in the total births for those years ( http://www.infoplease.com/ipa/A0005067.html ) you only get a 26% drop (2,559,000 * 0.07524 = 192539 vs 3,913,000 * 0.03645 = 142629).

    So, guessing the age by the name Ruth, you’re really asking “of people alive at this moment, which age group has the most Ruths” not “which year were the most Ruths born” or not “which year had the highest share of Ruths”.

  5. Brian is correct. The ‘learned-machine’ is predicting your likely age based on your name. 102 wouldn’t be a valid answer. That being said, I don’t know where the data came from, so who knows if it is accurate at all.

  6. Also, about Thom – I tried several different variations on names, and it didn’t make much of a difference. I bet it matches a name to some ‘root’ version of the name and uses that.

    As a side note, I tried Murphy and Max, and they were predicted to be in the neighborhood of 8 or 9, which is kind of funny.

Comments are closed.