Problem
Identifying a bird's species from a photo is a classic computer vision problem, but it is rarely obvious upfront which family of methods will handle it best on a given dataset. A classical statistical approach, a density based method, and a deep learning model can all be defended on paper.
Approach
Implemented and compared four distinct ML subdisciplines on the same bird image dataset: dimensionality reduction for visualizing and simplifying the feature space, traditional statistical classification models, density estimation methods, and convolutional neural networks. Built in Python using Sklearn, TensorFlow, and OpenCV, with the full methodology and findings written up in LaTeX.
Result
A side-by-side comparison of four fundamentally different modeling approaches on the same task, documented in a full technical report covering the trade-offs between them.