Contents

Description of demo_binaryclass_SVM.m

Demonstrates use of Support Vector Machines for binary classification with various loss functions. Support vectors are circled in red in plots below.

clear all
close all
generateData_slanted

usage of SVM binary classification

options_svm1 = [];
options_svm1.lambdaL2 = 0;
options_svm1.addBias = 1;
model_svm1 = ml_binaryclass_SVM(Xtrain, ytrain, options_svm1);
yhat_svm1 = model_svm1.predict(model_svm1, Xtest);
testError_svm1 = mean(yhat_svm1 ~= ytest);
fprintf('Averaged misclassification test error with %s is: %.3f\n', ...
        model_svm1.name, testError_svm1);
XtrainwBias = [ones(size(Xtrain, 1), 1), Xtrain];
figure;
plot2DClassifier(Xtrain, ytrain, model_svm1);
hold on;
indx = 1-ytrain.*(XtrainwBias*model_svm1.w) >= 0;
plot(Xtrain(indx,1),Xtrain(indx,2),'o','color','r');

Averaged misclassification test error with SVM Binary Classification is: 0.080

usage of SSVM binary classification

options_svm2 = [];
options_svm2.lambdaL2 = 0;
options_svm2.addBias = 1;
model_svm2 = ml_binaryclass_SSVM(Xtrain, ytrain, options_svm2);
yhat_svm2 = model_svm2.predict(model_svm2, Xtest);
testError_svm2 = mean(yhat_svm2 ~= ytest);
fprintf('Averaged misclassification test error with %s is: %.3f\n', ...
        model_svm2.name, testError_svm2);
figure;
plot2DClassifier(Xtrain, ytrain, model_svm2);
hold on;
indx = 1-ytrain.*(XtrainwBias*model_svm2.w) >= 0;
plot(Xtrain(indx,1),Xtrain(indx,2),'o','color','r');

Averaged misclassification test error with Squared Hinge Loss SVM Binary Classification is: 0.044

usage of HSVM binary classification with 0.2 epsilon

options_svm3 = [];
options_svm3.lambdaL2 = 0;
options_svm3.epsilon = 0.2;
options_svm3.addBias = 1;
model_svm3 = ml_binaryclass_HSVM(Xtrain, ytrain, options_svm3);
yhat_svm3 = model_svm3.predict(model_svm3, Xtest);
testError_svm3 = mean(yhat_svm3 ~= ytest);
fprintf('Averaged misclassification test error with %s is: %.3f\n', ...
        model_svm3.name, testError_svm3);
figure;
plot2DClassifier(Xtrain, ytrain, model_svm3);
hold on;
indx = 1-ytrain.*(XtrainwBias*model_svm3.w) >= 0;
plot(Xtrain(indx,1),Xtrain(indx,2),'o','color','r');

Averaged misclassification test error with Huberized Hinge SVM Binary Classification, 0.2 \epsilon is: 0.058

usage of HSVM binary classification with 0.8 epsilon

options_svm4 = [];
options_svm4.lambdaL2 = 0;
options_svm4.epsilon = 0.8;
options_svm4.addBias = 1;
model_svm4 = ml_binaryclass_HSVM(Xtrain, ytrain, options_svm4);
yhat_svm4 = model_svm4.predict(model_svm4, Xtest);
testError_svm4 = mean(yhat_svm4 ~= ytest);
fprintf('Averaged misclassification test error with %s is: %.3f\n', ...
        model_svm4.name, testError_svm4);
figure;
plot2DClassifier(Xtrain, ytrain, model_svm4);
hold on;
indx = 1-ytrain.*(XtrainwBias*model_svm4.w) >= 0;
plot(Xtrain(indx,1),Xtrain(indx,2),'o','color','r');

Averaged misclassification test error with Huberized Hinge SVM Binary Classification, 0.8 \epsilon is: 0.071

Published with MATLAB® R2015a