Efficient M-fold Cross-validation Algorithm for KNearest Neighbors

This project investigates m-fold cross-validation algorithms for automatic selection of k with k-nearest neighbors problems. An algorithm taxonomy is used to identify different m-fold cross-validation algorithms. The taxonomy contains three elements: the order of computing, the number of indexes and the number of threads. Different combinations of these elements produce different algorithms. Ten reasonable algorithms are implemented and tested on four datasets. These datasets are of different dimensions and different sizes. By analyzing the performance and results of these ten algorithms, the functionality of different elements in the taxonomy in different situation is identified.