Syuntai-lh / crossProductKernel-noisyData

Experiments using a support vector machine with kernels on fuzzy sets on noise data

Cross product kernels on fuzzy sets

Synopsis

Experiments using cross product kernels on fuzzy sets and SVM on PIMA attribute Noise data. Reference:

Guevara, Jorge and Canu, Stephane and Hirata Jr, Roberto Cross product kernels on fuzzy sets for fuzzy set similarity, 2017 IEEE International Conference on Fuzzy Systems - FUZZ-IEEE 2017

---

Prerequisites

• SVM-KM from http://asi.insa-rouen.fr/enseignants/~arakoto/toolbox/ (included)
• MATLAB 2013a for the experiments
• R for the plots
• All the twelve 5-fold versions of the PIMA and SONAR noise attribute data set from the Keel repository http://sci2s.ugr.es/keel/attributeNoise.php (pima-5an-nn, pima-10an-nn, etc). Put the datasets in separate folders within the datasets folder, i.e., /datasets/pima-5an-cn-5-fold/*.dat

Code Example

To run an experiment using the cross product kernels on fuzzy sets (linear, exponential and gaussian) usint the first fuzzification approach described in the paper (fuzz1) on the clean train - noise test (cn) version of the data with 15% level of noise level type in the MATLAB prompt:

      experiments('pima' , 15, 'cn','fuzz1')

To run all the experiments for the PIMA attribute noisy datasets using screen run this:

 screen  -d -m matlab -nodisplay -nosplash -r    "experiments('pima' , 5, 'cn','crisp')"
 screen  -d -m matlab -nodisplay -nosplash -r    "experiments('pima' , 10, 'cn','crisp')"
 screen  -d -m matlab -nodisplay -nosplash -r    "experiments('pima' , 15, 'cn','crisp')"
 screen  -d -m matlab -nodisplay -nosplash -r    "experiments('pima' , 20, 'cn','crisp')"
 screen  -d -m matlab -nodisplay -nosplash -r    "experiments('pima' , 5, 'cn','fuzz1')"
 screen  -d -m matlab -nodisplay -nosplash -r    "experiments('pima' , 10, 'cn','fuzz1')"
 screen  -d -m matlab -nodisplay -nosplash -r    "experiments('pima' , 15, 'cn','fuzz1')"
 screen  -d -m matlab -nodisplay -nosplash -r    "experiments('pima' , 20, 'cn','fuzz1')"
 screen  -d -m matlab -nodisplay -nosplash -r    "experiments('pima' , 5, 'cn','fuzz2')"
 screen  -d -m matlab -nodisplay -nosplash -r    "experiments('pima' , 10, 'cn','fuzz2')"
 screen  -d -m matlab -nodisplay -nosplash -r    "experiments('pima' , 15, 'cn','fuzz2')"
 screen  -d -m matlab -nodisplay -nosplash -r    "experiments('pima' , 20, 'cn','fuzz2')"

Generating a cvs file with the results

This generate a csv with all the results. Run the MATLAB script:

testResultsIntoCVS

Generating some plots

Run the R script

getPlots

Run the Matlab script

generateFigures

Motivation

This project is part of my research on kernels on fuzzy sets

Contributor

[e-mail]