Year: 2017
Type: dissertation
Abstract: In this thesis, for the purpose of utilising large-scale data, we develop general purpose technologies that expose features and problems of such large-scale data generated in various environments.In detail, we first propose a new anomaly detection method for various types of time series data for the purpose of basic data processing.The proposed anomaly detection method is an interval segmentation method assuming that not only short-term anomalies (bursts) but also long-term anomalies (latent changes) are mixed.As an application of this method, we evaluated time-series data in consideration of temporal reliability using the segmentation result of the method.Subsequently, based on the time-series data evaluation method by the proposed interval segmentation method, we construct an evaluation model of data categories introducing a temporal decay function which is another approach to temporal reliability and a spatial decay function related to the position information.Furthermore, in order to expand the evaluation method as macroscopic analysis, we propose a complete network model in which the data objects are nodes and clarify the features of the data categories from the various node evaluations.Finally, as an application of the proposed network model, we also dealt with time series analysis when the nodes are the data categories.From the results of comparative experiments using representative conventional methods and naive methods for these problems, we confirmed the effectiveness and usefulness of the proposed method for time-series data with multiple attributes, spatiotemporal data with position and timestamps, and data acquired by batch processes.
Author: 山岸, 祐己
Language: Japanese
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7145583591
