Mining And Applications Of The Temporal Approximate Periodicity Based On Multi-Attribute Characteristics

With the development of temporal data mining, periodicity mining, as a very meanful characteristics for understanding temporal data and predicting future trends,has now become a very important research topic. However, very few people study temporal approximate periodicity and association rules based on multi-attribute characteristics. Multi-dimensional temporal approximate periodicity refers a multi-attribute incident after a time period (, which the time cycle can be fixed within the time scope fluctuations), then this incident takes place again. The association rules still research the extent of effects of ex-piece rule to the rule behind. It is a very significant that the same object often exist mutual influence in periodicity between the various attributes and the different objects also influence each other periodically for the mutual competition or cooperative relations. The periodicity is often manifested as an approximate periodicity. If we apply approximate cycle and approximate periodical association rules to financial, meteorological, hydrological, medical care, in areas such as supermarkets,, it is considerable practical value to solve the management and decision-making problems .The paper studies the following aspects of temporal approximate periodicity.(1) We introduce temporal periodicity mining research background, deficiencies and the content of this paper.(2) We give some definitions of type of temporal, temporal factor and temporal particle size for multi-dimensional events, and temporal prediction model, as well as multi-dimensional temporal association rules perodicity model.(3) We propose mathematical definition of multi-dimensional approximate periodicity, multi-dimensional approximation periodicity model with support and confidence, approximate accuracy and coverage model of approximate periodicity. Then we prove the relevant property. At the same time, we study the algorithm of multi-dimensional approximate periodicity based on the temporal database and two kinds of cluster methods as hierarchical cluster and SOM cluster. Experiment results show that the algorithm is effective.(4) We propose the concepts and properties of three kinds of temporal multidimensional approximate peroidicity association rules, and study the algorithm based on SQL language and SOM cluster. At the same time, we discuss prediction of the association rules by the multi-dimensional approximate perodicity association rules.The main results in this paper are as follows. First, we give the multi-dimensional events, multi-dimensional expansion of the temporal prediction model, and then propose multi-dimensional cycle association rules model. Second, we give a multi-dimensional temporal approximate perodicity mathematical model and minging algorithms based on temporal database and Hierarchical cluster and SOM cluster. Finnaly, we persent three different types of multi-dimensional temporal approximate perodicity association rules and their algorithms based on the temporal database and SOM cluster to predict the practice .