ASSORTED PERIODIC PATTERNS IN TIME SERIES DATABASE USING MINING
Associate Professor, Dept. of IT, Bharath University, Chennai -6000731
|Related article at Pubmed, Scholar Google|
Time series is a collection of data values which is gathered at uniform intervals of time to reflect certain behavior of an object. Time series has three types of periodic patterns which are Symbol periodicity, Sequence periodicity, Segment periodicity (S.S.S).Mining periodic pattern has number of applications such as prediction and forecasting. We use time series data base mining in super stores, network delays, power consumption. The data which we observe in the periodic patterns may contain noise .So there is a need to analyze the whole time series data or subset of it which can effectively handle different type of noise(unwanted data).Here we will use an algorithm ,which uses suffix tree as a basic data structure .The algorithm is divided into two phases. In the first phase we build a suffix tree for time series and in the second phase we use the same constructed suffix tree to calculate the periodicity of various patterns in time series. Another important aspect of this algorithm is redundant period pruning, i.e we ignore a redundant period ahead of time as it does not waste time to investigate a period which is already redundant thus proving more efficient. In the end we calculate the confidence of the periodic pattern of time series and depending on the value of confidence (0<=p<=1) we rate the time series.