|Title||Learning Distance Functions for KNN Classifiers|
|Related Course(s)||Advanced Data Mining, Learning from Data|
|Description||k-Nearest Neighbor (KNN) is a popular non-parametric classifier.
To predict the class of a new point, we find the k nearest points, and use their known class values to predict the unknown class of the new point.
In doing so, we have to choose some distance function to determine how "near" two points are to each other.
The objective of this project is to learn the appropriate distance function from the data as well.
We consider a class of distance functions with adjustable parameters, and try to determine the parameter values that give the best results in prediction.
The algorithm is implemented in the R language, and experimentally tested on data sets from the UCI Machine Learning repository.