23-01-2010, 08:15 PM
[attachment=1397]
Abstract
A model for identifying similar time series has beendeveloped. Two time series are considered similar ifthey have enough non overlapping time ordered subsequencesthat are similar. The two subsequencesare considered to be similar if one is enclosed withinan envelope of a user defined width around another.Amplitude of one of the two sequences is scaled by asuitable amount to make the matching scale invariant.We also allow a small gap between the matchingsubsequences. A static index structure namely kdtreeis used to index the time series and the Rangesearch methodology is used to search for the similarsubsequences. The highlight of this model is thatthe similar subsequences are not stored repeatedlyin the index structure, but only the time stamp correspondingto it is updated in the external node ofthe kd-tree.Interesting results are found to show thevalidity of the model.