One of the main problems when machine-learning classifiers are employed in practice is to determine whether classifications assigned to new instances are reliable. Version Space Support Vector Machines are one of the most successful approaches to this problem. They maintain version spaces in hypothesis spaces of oriented hyperplanes. The classification rule of unanimous voting is realized using Support Vector Machines. It has been shown that when the hypothesis space contains (close approximations of) the target classifiers, the unanimous-voting classification rule applied on the version spaces guarantees that the classification assigned to each instance is 100% reliable.

The experiments with Version Space Support Vector Machines have shown that it is difficult to apply them for multi-class problems (datasets contain more then two classes). To overcome this obstacle we propose two research projects. Each project can be a base for a Bachelor project with a possibility to be continued later as a Master project.

The one-against-all approach is widely known approach to convert a multi-class problem into a series of binary problems. To be more precise, given N number of classes, the approach generates N number of binary problems so that each K-th binary problem considers the instances of the class K as positive and the instances of the remaining instances as negative. For each binary problem a binary classifier is built. Thus, each multi class problem can be solved using a binary classifier.

Version Space Support Vector Machines are binary classifiers. The main objective of this research project is to apply them to multi-class problems using the one-against-all approach.

The project is experimental. First, a system of the Version Space Support Vector Machines using the one-against-all approach has to be implemented in the WEKA data-miming environment. Once the system has been implemented, a series of experiments on standard datasets has to be performed.

The pair-wise approach is widely known approach to convert a multi-class problem into a series of binary problems. To be more precise, given N number of classes, the approach generates N(N-1)/2 number of binary problems so that each binary problem includes labeled instances of exactly two classes. For each binary problem a binary classifier is built. Thus, each multi class problem can be solved using a binary classifier.

Version Space Support Vector Machines are binary classifiers. The main objective of this research project is to apply them to multi-class problems using the pair-wise approach.

The project is experimental. First, a system of the Version Space Support Vector Machines using the pair-wise approach has to be implemented in the WEKA data-miming environment. Once the system has been implemented, a series of experiments on standard datasets has to be performed.