Time period:
Start Date: December, 2019End Date: January, 2020
Background
This was my final project for a course I did on Data Science. The attempt I made was futile, but it is a good reference to see how I thought stuff worked and tried to immaturely combine features for some miracle to happen.
The Problem Statement
I downloaded a huge celebrity-images dataset and decided to build a Model that could tell the males apart from the females. In the end, the best model I could come up with had 60% accuracy, which is basically lucky guessing.
The Process
I first took all the images, used a facial recognition model to draw rectangles around the faces of everyone, cropped the images and turned them all black and white. Then I treated each and every image as an array of pixels, make this rectangular 2-D array into a 1-D list and treated each pixel as a different column.
Then I tried to run different models, by having over 16000 input columns and one output column, and passing it all through Random Forest, K-Means and Logistic Regression models. Logistic Regression worked best, and had approximately 60% accuracy. Other models stayed at the 50-52% range which meant they had not even noticed a pattern and were just guessing.
Resources
Links
This is a link to the Celeb-A Faces dataset that I used for my face classification project.