Efficient Classification Method for Large Dataset

Clustering analysis is used to explore the classification for large dataset and Canberra distance is generalized so that it can process the data with categorical attributes. Based on the generalized Canberra distance definition, an instance of constraint-based clustering is introduced. Meanwhile, the nearest neighbor classification is improved. Class-labeled clusters are regarded as classifying models used for classifying data. The proposed classification method can discover the data of big difference from the instances in training data, which may mean a new data type. The theoretic analysis shows that the learning process and the classifying process of the proposed classification method has nearly linear time complexity, which makes the method results in good scalability and applicable to large dataset. The experimental results demonstrate that our method is effective and practicable, and has high prediction accuracy