Time-series Data Mining Based On Rough Sets And Its Applications

Time series data mining is the search for underlying and useful information in large volumes of time-series data set and using the information to predict the time-series future. Furthermore, more and more people began to pay more attention to time series data mining, it is a new research hotspot among data mining field. Nowadays, the study about time-series mining mainly focus on trend analysis ,similarity serching ,time sequence query and rules discovery. In this dissertation, the research of the time series data mining technique and its application in forecasting the stock market, taken as a part of the projections of advisor. In this thesis, classic forecast models are firstly devoted to discussing and researching. Futhermore, tme-series data mining model and technique based on Rough Sets which is a strong knowledge retrieve tools are proposed. The Application of time-series mining technique in stock closing price prediction is reserched. The main work and the primary results and hard core of this dissertation are summarized as following:1.In the research of secondary exponential smoothing modle forecast model, due to determining problem of smoothing parameter, a method of determing the secondary exponential smoothing parameter based on OWA is proposed. It is different from artificial determinations and automatically assignation which considered historical error and prediction error. It inherits merits of above mentioned methods.Optimal smoothing parameter value which suitable to a certain prediction system is obtained according to OWA aggreation operator thereby reach the better prediction results.2. Trending variation ratio structure sequence is defined by author based on trending structure sequence. The author regards the latest time sub-series as the information collection of the time series. Basing on this point of view, fuzzy linguistic value discription on various trending variation ratio are given,e.g. ascend_rapid, ascend_moderate, ascend_low, descend_rapid, equal, descend_moderate,descend_low.3.Time-series mining approach based on rough sets and trending variation ratio structure sequence is brough forward. The approach is that the time series set waiting for mining is first converted into its trending variation ratio structure series set, then the trending variation ratio structure series set is pre-processed by fuzzy discrete method ,and then the trending variation ratio structure series set is transformed into a normal time-series sample set with the "mobile window mthod". Finally, the normal time-series sample set is imported into a special decision table where the last object of the decision table is the latest time sub-series immediately following the forecasting goal of which is placed in the decision attribute posion,and the cardinal number of the condition attribute set is equal to the lengthe of the latest time sub-series. And the forecasting goal is predicted and reasoned out with fuzzy reasoning and pattern match methods.