关键词: 分类, 随机性, 概率数据, 辨识距离

Abstract: Classification as an important problem in data mining is widely studied and applied nowadays, but the previous study is mainly about classification on certain data. Since probabilistic data exist and are widely used in many fields, such as sensor data, it is necessary to do feature selection for probabilistic databases. Firstly, this paper proposes a new probabilistic data model, which considers not only the randomness but also the similarity of different intervals. Secondly, in order to do classification for such probabilistic data, this paper designs a discernible distance to measure the distance between such tuples. Finally, this paper proposes a basic rule-based classification algorithm, and develops a new variable distance to reduce classification sensitivity to noise or perturbation. The Experimental results verify the effectiveness of the proposed algorithm.

Key words: classification, randomness, probabilistic data, discernible distance