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.
A decision boundary was also generated:
Created with Radu's Machine Learning Course and samples generated by his students