Framework for Performance Comparison of Classifiers

Abstract

Data stream mining is the process of extracting knowledge structures from continuous, rapid data records. Classification is one of the task involved in data stream mining that maps data into predefined groups or classes. Most of the stream learning algorithms learn decision models that continuously evolve over time, run in resource-aware environments, detect and react to changes in the environment generating data. Built model will always have high accuracy on the training data, but performance on unseen data is to be checked. Performance of different classifiers for same task in same environment can differ, so there is need for some method which will help one to select the best suited classifier for the required task. Performance comparison will be effective if graphical interface facility is given. There is a need of user friendly interface having facility of multiple classifier selection for performance comparison, saving environment for future use and plotting the performance graph of classifiers. A framework which will provide different measures for performance comparison like true positive rate, true negative rate etc. is today’s requirement. Objective of this paper is to enhance the existing software used for stream data analysis with the above mentioned facilities.