MR- ANT MINER SYSTEM FOR THE DATA STREAM CLASSIFICATION

Abstract

Data stream classification plays an important concept in data mining. The emerge of new data stream and concept drifts are the two challenges found in the traditional knowledge discovering methods. In this paper, we introduced the mr-Ant system to evaluate the drifts from a data stream. Our ant system exploits an effective global search and discovers the refreshed classes from the past learning models in a parallelized manner. Consequently, it saves the features space by lessening its space and time complexity. A pheromone classifier based on Principal Component Analysis (PCA) has been introduced to measure the performances such as accuracy of detecting new classes, precision of relevant classes, error rate of mislabeled class instances and time taken for new class detection which proves that our proposed work was the best.