Feature Evolving Data Streams using SVM Kernel in the Multi Novel Class Detection
Chaitrali Chavan, Vinod S. Wadne · International Journal of Computer Applications · 2015
There are many challenges which community faces In Data Mining, concerning with the data stream categorization.The four different issue of categorization viz.infinite length, concept drift, concept, development feature, development.Due to infinite length of data, it is impossible to store and use the traditional data.Many researchers focus on the issues of all of the four challenges for data stream categorization.In this system novel class are detected by using the Gini coefficient method and outliers are detected by using the adaptive threshold method.We used SVM method for detecting the multi novel class detection.In the present system data are divided into fixed sized of chunks for classifying the stream instances, because of this system fail to capture the concept drift immediately.That's why solution of this method to the change point detection method which are trying to determine the chunk size dynamically on the data stream.The computational complexity is improved by clustering algorithm the performance is checked of the system by using the forest outlier dataset.