presentation.Rpres

Effect of the number of classes on accuracy of classification
========================================================
author: Vladimir Metelitsa
date: 19/01/2017
autosize: true

Unsupervised and Supervised Classification
========================================================

We have gone over unsupervised and supervised classification in class.
Quick refresher:
- Unsupervised: Model "guesses" classes itself
- Supervised: You provide training data for the model as a vector file

What I made
========================================================

- Web interface to play with different inputs into classification
- Support for both unsupervised and supervised classification
- Customisable number of classes and sample size
- Display of accuracy for supervised classification

Demo time =)
========================================================



Results
========================================================

```{r, echo=FALSE}
classes <- c(2, 3, 4, 5, 6, 7, 8, 9, 10)
# precalculated using the shiny script, inserted as values to save processing time
accuracies <- c(100, 100, 99.5762711864407, 96.8956406869221, 87.5982532751092, 86.1079723791588, 84.5211581291759, 82.8207051762941, 79.203187250996)

plot.default(classes, accuracies)
abline(lm(accuracies~classes), col="blue")
```

Clearly less accuracy the more classes we have.

But why?

Try it at home
========================================================

If you want to play with this, you can download it from my Github repository at: 

https://github.com/Green-Cat/M15RemoteSensingProject

Or try it online at a URL which will be added later to the description of my repository.