Big data classification based on multi-view method

The traditional classification technique cannot be applied directly to the big data problem mainly due to the high time complexity issue. The existing method trained a classifier using the dataset simplified by the clustering approach. However, the main drawback of this method is that the dataset may be over- or under-simplified. In this paper, we revises this method for a big data classification task. Our proposed method considers multi-view rather than single training set. Several simplified datasets are firstly generated based on the clustering technique with different radii. Then several classifiers are trained on these datasets and are combined together based on the boosting method to form a multiple classifier system (MCS). Therefore, different levels of the dataset can be learnt, and also the time complexity of the classification method is practicable. The experimental result suggests that our method achieves better performance in comparison to the traditional SVM method on the big data problem.