Machine Learning Demo
A model that uses kNN-classification to categorize hand-drawn images into eight different categories.
Includes webpages to:
- generate and save new samples locally
- Display samples, view decision boundaries, and inspect individual samples on a graph
Model trained with NodeJS, webpages created with HTML/CSS and JavaScript
The parameters used were each images' height and width. Even using this basic trait and a few hundred images, the model was able to reach over 40% accuracy compared to random selection(12.5%).
Images were divided into eight categories: car, fish, house, tree, bicycle, guitar, pencil, and clock. The unknown images were categorized by selecting the closest 50 neighboring drawings and picking the highest occurence.
Created with Radu's Machine Learning Course and samples generated by his students