Affiliations: School of Computer Science and Engineering, North Minzu University, Yinchuan, China
Correspondence:
[*]
Corresponding author. Meng Han, School of Computer Science and Engineering, North Minzu University, Yinchuan 750021, China. Tel.: +86 09512067872; E-mail: [email protected].
Abstract: Time decay model (TDM) is frequently used for mining frequent patterns on data streams, because the information embedded in the data from the new transactions is particularly valuable. However, some existing methods on designing decay factor of TDM are random, so their results are unsteady. Some other methods focus on only 100% recall or 100% precision of algorithm, but the corresponding high precision or high recall is ignored. In order to balance high recall and high precision of algorithm, meanwhile, ensure the stability of the result, a novel average decay factor is designed. In addition, to further increase the weights of the latest transactions and reduce the weights of historical transactions, another novel Gaussian decay factor is proposed. Hence, based on an analysis of existing decay factors, this paper aims to design two novel decay factors and two novel TDMs. Algorithms based on these two TDMs are designed to discover frequent patterns over data streams. The methods of mining frequent patterns on high density or low density data streams are evaluated via experiments. This paper’s research findings show that the application of average time decay factor can balance the high recall and high precision of algorithm. And Gaussian decay factor can produce better performance than existing algorithms.
Keywords: Data streams mining, frequent pattern mining, decay factor, time decay model, Gaussian function
