一种基于趋势分形维数的股指时间序列相似性分析方法

为了提高股指时间序列相似性分析的准确性, 提出趋势分形维数的概念, 并基于此定义了相似性分析方法. 趋势分形维数包含阳线维和阴线维, 能更好地反映市场跌涨变化趋势, 基于该维数的相似性度量方法能够提高相似性度量的准确性. 通过与其他两种相似性度量方法对比, 进一步说明该方法的优越性.     

在线挖掘数据流混合窗口中闭频繁项集

在数据流挖掘中,界标窗体考虑了历史模式对当前挖掘的影响,但没考虑到随时间的推移模式衰减的问题。滑动窗口能记录最新、最有用的模式,但窗口的最佳大小无法准确确定。针对一些仿真系统中具有数据流特点的数据,提出了一种挖掘混合窗口中闭频繁项集的方法T-Moment。该方法能在单遍扫描数据流的条件下完整地记录模式信息。同时,T-Moment提出的减枝方法能很好地降低滑动窗口树F-tree的空间复杂度与闭频繁模式树T-tree的维护代价。此外,该方法提出的时间衰减机制能区分历史和最新模式。大量仿真实验结果表明,T-Moment有很好的效率和准确性。

Abstract：

In data mining,boundary window considers the influence of history pattern to the current mining result,but do not think over mode decaying as time passed. Sliding window can record the latest and most useful patterns,but the best size can not be accurately determined. To aim at data with the characteristics of data flow in some simulation systems,a method for mining the closed frequent patterns in the mixed window of data stream was proposed. The pattern of data stream could be completely recorded by scanning the stream only once. And the pruning method of T-Moment could reduce the space complexity of sliding window tree and the maintenance cost of the closed frequent patterns tree. To differentiate the historical and the latest patterns,a time decaying model was applied. The experimental results show that the algorithm has good efficiency and accuracy.