The Study Of Cluster And Association Rule Mining In Time Series

The rapid development of information technology brings huge data, and it is important how to mine useful knowledge in these data. The data mining seems to be a powerful tool to deal with this problem. Among these mass data, one of the important data types is time series (TS) which is data set that the data records are arranged over time. The various data produced in many files share this characteristic. In recent years, the research on TS has been one of the popular topics in data mining because TS can describe exactly the event change procedure over time. The dynamicity, high dimensions, characteristic dependency and noise are the unique structure of TS. For this reason, it is difficult to mine knowledge in TS by using of classical methods and algorithms. Therefore, researching on data mining in TS is a challenging and valuable problem in theory and practice.Many systems in the world can be modeled by complex network. A typical network is composed of nodes and edges connected nodes, where nodes represent the different entities and edges reflect the relations between a pair of nodes. Recently, researches pay more and more attention to complex network which are produced in sciences, engineering technology and even society. One of the important characteristics in complex network is community structure. Many methods or algorithms have been proposed to deal with how to detect community structures in complex network. In fact, searching for communities in network is equivalent to clustering data. It is worthwhile to study how to clustering TS in terms of complex network.The cluster and associate rule mining are two basic problems in data mining. By weighting time in TS and complex network, the following work has been done in this dissertation.An algorithm for time series clustering based on the spectral bisection method of Normal matrix is proposed. The algorithm firstly transforms time series data into vector forms, calculates the similarity between any pairs of time series and constructs complex network. Then the complex network will be divided into communities by using of the method of Normal matrix. The time series are clustered in terms of the results of partitioning network. Finally, in order to verify the feasibility and effectiveness of the presented method, the real stock time series are analyzed, other methods are compared on two real datasets and the desired results are obtained.A method for mining association rules with weight based on context is presented by introducing a weighted formula. The records are weighted according to time under the guidance of the recent-biased principle. The algorithm extracts association rules from the time-place context in which users are interested. Compared to the traditional methods, the relevant information with different granularity levels can be analyzed from lower granularity level to higher one in terms of mapping sequence of the concept. The approach is of advantage to policy maker for making more accurate decision analysis and more optimal strategies.