Skip to content
MATLAB implementation of A Feature Selection Based on Perturbation Theory
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
Data
PFS.m
README.md
cAccInner.m
cAccOutter.m
readLargeCSV.m

README.md

title authors date
PerturbationFeatureSelection
Javad Rahimipour Anaraki and Hamid Usefi
29/05/18

Notice

The code PFS.m is available now and a link to the paper will be provided soon. If you need more details and explanation about the algorithm, please contact Javad Rahimipour Anaraki or Hamid Usefi.

Use case

To determine the most important features using the algorithm described in "A Feature Selection based on Perturbation Theory" by Javad Rahimipour Anaraki and Hamid Usefi

Here is two links to the paper: arXiv and Expert Systems With Applications

Compile

This code can be run using MATLAB R2006a and above

Run

To run the code, open PFS.m and choose a dataset to apply the method to. The code strats reading the selected dataset using readLargeCSV.m written by Cedric Wannaz. Then it selects the most important features and find the best subset by looking at the classification accuracies returned by cAcc.m divided by the size of seleted subsets. Finally, a subset with the best accuracy and the smallest number of features is selected and returned as the output. All datasets are stored in Data folder and originally adopted from UCI Machine Learning Repository and ASU feature selection datasets

Note

  • In order to get accuracy using decision tree, support vector machine or k-nearest neighbour the corresponding line in the cAccInner.m should be uncommented
  • Datasets should have no column and/or row names, and the class values should be all numeric
You can’t perform that action at this time.