Posted by Danny TarlowThis is a guest post by Rolf, the driving force behind our early co-leader, "My Robots Wiked Smaht". After three games, they are one of only two teams to have a perfect 3 of 3 bracket.
Entry Name: My Robots Wicked Smaht
Team Member Names: Rolf and Andrew
Description: Our backgrounds are more on the human learning side of things, so we took a fairly simple approach to creating a bracket picking robot. Our robot uses a simple regression to identify key variables in order to enhance our variant on RPI rankings.
As an early leader in the March Madness Predictive Analytics Challenge, Danny suggested that I write a guest blog post detailing some of the finer points of my prediction algorithm - "My Robots Wicked Smaht." First off, I must mention that although I am a regular reader of "This Number Crunching Life," my formal mathematics and computer science background is pretty limited. I'm currently doing a PhD in Education, so I've been more focused on human learning than machine learning. But enough chit chat.
My first task in creating this robot was to learn how to use excel. That was the trickiest part, actually. Once I had organized the 2009 data into some statistics that I thought might be relevant, I consulted with my colleague Andrew. He then ran a regression and determined that wins in January and February were more highly correlated with wins in march (our goal) than those from November or December. Taking this information, we created a kind of modified RPI ranking ( had to look this one up as well), with extra weight placed on late season wins in addition to the standard factors of winning percentage, opponent's winning percentage, and opponent's opponent's winning percentage.
I was hoping to also factor in the average team height, but that seemed like a whole lot to do. Next Year!