A proximate dynamics model for data mining

摘要

Many association rules with low supports and high confidence are commonly not convincing, so how to enhance the conviction of such rules is a big issue. In this paper, we explore the dynamics features of the domain dataset, and by predefining some dynamics parameters (as priori knowledge), we construct a proximate dynamics model for data mining so as to enhance the conviction of the rules with low supports. For constructing such dynamics model for data mining, we adopt three techniques: (1) a large domain dataset is classified into several sub-clusters, which we regard as proximate dynamics systems; (2) the data mining process is a solving process of differential equations, which captures the changes of the data, not only the values themselves; and (3) a weighting method is used to synthesize the local mining results with the users’ preferences. Although we “arbitrarily” apply the dynamics parameters into the sub-clusters of the given dataset, the experimental results are very well by comparing with FP-growth algorithm and CLOSET+ algorithm. Experiments conducted on the distributed network with three real life datasets show that the proposed method can discover the knowledge based on dynamics, which is potentially useful for mining the rules with low supports and high confidence.