Purpose:
The purpose of this project is to design a multi-class classifier using linear programming approach to construct SVMs. It includes two approaches: One vs One method and One vs All method. The accuracy to train a set of 400-feature, 10-class dataset is approximately 86.6%.
Files:
- SeparatingHyperplane.m helps creating the seperating hyperplane for two sets of labeled data using cvxopt package
- MyClassifier.m trains sets of labeled data to construct a classifier to predict the future unlabeled dataset.