One class classification using fitcsvm

Classifier<- svm(PhaseI_Data,type='one-classification',kernel = "radial",

nu= nu,gamma= gamma, scale = F)

Above written code is from R. where nu is penalty factor, gamma is kernel bandwidth in radial kernel.

Can someone help me to write equivalent code in matlab.

I have tried multiple times but it is giving different results in both.

Answer 1

[Hitesh]

Hi Nainsi,

MATLAB uses the "fitcsvm function". However, "fitcsvm"does not support one-class classification directly. You need to use the "fitcsvm" with 'OutlierFraction', which is equivalent to 1 - nu in R.

Kindly refer to the following code as example to implement "One class classification using fitcsvm".

% Assume PhaseI_Data is an n-by-d matrix (observations x features)
% and already normalized/scaled similarly to how it was used in R.
% Example dummy data: 100 points in 2D

PhaseI_Data = randn(100, 2);  % Normally distributed data

% Convert R gamma to MATLAB's kernel scale:
% R: k(x, y) = exp(-gamma * ||x - y||^2)
% MATLAB: uses 'KernelScale', which is sigma such that
%         k(x, y) = exp(-||x - y||^2 / (2*sigma^2))

gamma = 0.1;  % or whatever value you used in R
sigma = sqrt(1 / (2 * gamma));

% MATLAB One-Class SVM equivalent:
SVMModel = fitcsvm(PhaseI_Data, ones(size(PhaseI_Data,1),1), ...
    'KernelFunction', 'rbf', ...
    'KernelScale', sigma, ...
    'OutlierFraction', 1 - 0.1, ...
    'Standardize', false);

For more information regarding "fitcsvm", kindly refer to the following MATLAB documentation:

https://www.mathworks.com/help/stats/fitcsvm.html