Unsupervised Learning of Religious Facial Features

Posted by Christopher Mertin on April 29, 2017 in Project • 1 min read

This was my semester project at the University of Utah for my Data Mining class. The goal was to come up with a project such that we had to mine our own data and use some basic unsupervised methods on the data.

Based on a paper published in a psychology journal, I decided to test to see if a computer could differentiate between mormons and non-mormons with unsupervised learning techniques. My report on the project here, which states all of the technical details and the algorithms that I utilized, with the poster that I had to present being located at this url. This project also won the “best project” award out of all the other 50+ projects that were being presented.

In the psychology journal publication, they used randomly sampled people to see if people could distinguish the differences between being mormon and non-mormon and achieved a 55% accuracy.

I set out to obtain my own data with the help of the Tinder API, and I utilized various techniques such as SIFT features, Eigenfaces, and clustering algorithms to see if a computer could learn to distinguish between the two groups. This resulted in a 59% accuracy for the clustering algorithms, and up to 80% accuracy with eigenfaces.

The code and data that I used in this project can be found on my github.