A New Voting Method to Novel Class Detection Using Hoeffding Option Tree

Abstract: Data Stream mining is a process of extracting knowledge structure from continuous and rapid data records. Data stream size is extremely large. It’s a continuous flow of data So major problem related to data stream is infinite length, concept evolution and concept drift. Novel class detection is very interesting topic in a data stream mining. We can detect novel class using classification and clustering. Currently mostly uses decision tree using classification. In that Hoeffding Tree is better for data stream mining. And Hoeffding Option Tree is more better than Hoeffding Tree. In our paper we can use new voting method which is different than HOTDC. In our method we can use classification . So its supervised learning. Here classes are fixed before examined data. But when continuous data come then not all data are classified. Some data are misclassified. And this class is not in universal existing class then its known is a novel class. In our method when this type of class is detected then model is trained. So no require to collect those type of data. So when again this type of instance is come then its classified in that class. Using that method we can release from problem concept drift, concept evolution and infinite length.

Keywords: Novel Class, Hoeffding Option Tree, Outliers, Recurring Class, Hoeffding Bound