At the Knowledge Systems Group at NUST, one of the main areas of interest is currently learning from examples. In this context, we were given the assignment shown below:
Work is currently being done in the Knowledge Systems group on the problem of learning from examples, that is, given some input data, to induce rules that cover the examples. More specifically, we are interested in the generation of default rules (in the framework of Poole's Theorist system) from a set of primitive input data (typically, descriptions of objects and their properties), using the Rough Set approach.An algorithm created by Torulf Mollestad is available for creating default rules, and has been implemented by Jon Petter Hjulstad. The task will be to explore the possibilities this gives, and consider what more should be done to create a complete and user friendly system which utilizes the default rules. Finally, the project may include some testing of the system and of the usefulness of the default rules compared to other classification systems as for instance RSES.
We understood our task as mainly focusing on making a usable data mining system which creates default rules and is capable of using these. In order to do so, Rough Sets and data mining had to be studied. The program for creating default rules should be able to handle larger amounts of data than it actually could, and a classification system had to be built. Therefore, implementation issues has been an important part of our work.