The Analysis On Time Series Data Based Upon Rough Set Theory

This thesis discusses the method of obtaining time series information system, which is the base of converting into non-time series information system. However, few literatures involve the method of obtaining time series information system formerly. We can win time series information system from both origin discrete data and real-time time series information system.The aspect of non-time series information system reducing attributes is a key process before data mining. The strategy of reducing attribute based on discernable matrix is usually used among many methods. However, it has some insufficiencies including high computing complexity and low computing efficiency. Aim at that difficulty, the thesis provides an improved method under the illumination of some examples, which time complexity and space complexity is less than that based on discernable matrix and has the same effect.This dissertation also researches on a strategy of reducing attribute based upon conditional information entropy, which takes a connotative characteristic of time in attribute of non-time decision table converted from time decision table. This dissertation puts up with a tentative about time significance, which is used into the above strategy.Finally, we have a study on the acquisition of rules. Before providing an improved strategy on rule acquisition under the illumination of examples, we analyze a drawback of traditional one on gaining minimal rules set directly. The improved strategy can obtain minimal rules set directly.Using a database related with time series in the UCI database sets and a simulated program compiled by author, we validate the effect of the above strategies. The program can reduce attributes and obtain rules from time series data.