MapReduce Based Frameworks for Classifying Evolving Data Stream

Data Stream mining has some inherent challenges which are not present in traditional data mining. Stream data classification is a challenging problem because of two important properties: its infinite length and evolving nature. In our current work, we have proposed a multi-tiered ensemble based method to address these challenges of Data Stream mining. Though, it is a fast and robust method, it needs to build large number of AdaBoost ensembles for each numeric feature after receiving each new data chunk. Thus, it can face scalability issue especially when size of the data chunk is large or the number of numeric attributes is high. To address this problem, we propose two different approaches to form these large number of AdaBoost ensembles with MapReduce based parallelism. Each of these approaches help our base method to achieve significant scalability without compromising classification accuracy. We compare these approaches from different aspects of design. We also demonstrate and compare performance of the approaches on benchmark datasets in terms of execution time, speedup and scale up.