Indexing And Querying Of Time Series In Data Mining

Indexing and querying are fundamental problems in data mining. This dissertation addresses the problem of indexing and querying for time series, and discusses the related problems of the query methods, representation and similarity measures of time series. By making use of the proximity query method in computational geometry, the whole matching query, pattern query, inverse query and outlier detection in time series are studied. The main results of this dissertation are as follows:1. Query methods of time seriesAccording to the components of time series, three categories of querying are proposed, including point querying, pattern querying, and time series querying. According to proximity definition in computational geometry, the querying of time series has been expanded to range query, proximity query, and closest pair query. The development of time series querying can provide effective techniques for data mining.2. Representation and indexing of time seriesBased on the important point segmenting of time series, a new similarity measure by slope and length are developed. For the indexing of time series, the Voronoi diagram has been used for the organization of time series, and a new indexing structure named IC-indexing is proposed for the inverse query of time series.3. Similarity querying in time seriesWith the KL similarity measure, a new algorithm for whole matching of time series has been developed. The pattern querying problem is solved in optimal time based on the Voronoi diagram.4. Outlier detection in time seriesBy making use of the Voronoi diagram, a new density-based outlier detection algorithm is proposed, which runs in computation as O(nlogn), and has been used in the liner pattern outlier detection in time series.