Solving Multi-Class Classification Problem Using Support Vector Machine

Many data mining and pattern recognition tasks have a multi-class classification problem at their core. To give satisfactory results in operational contexts, certain applications necessitate high-end machine learning systems. In the case of supervised learning problems, the most efficient ones, such as Boosting, are mono-class, which brings the dilemma of converting a global multi-class problem into many binary issues while still being able to deliver an answer to the original multi-class problem at the end. They work for binary classifications well but sometimes it is necessary to have multi-classification. The current work intends to solve this multi-class challenge by offering a comprehensive framework that includes multiple multi-class classifiers as well as principal component analysis for feature selection. Their performances in terms of accuracy are tested on world-class benchmark data-sets. The data sets with appropriate training and test samples are used for experimentation. It is seen that linear SVM performs well among all the classifiers.