Year: 2013
Type: dissertation
Abstract: Finnish summary Diss.This work is devoted to the study of knowledge discovery using diffusion maps.Data manipulation and storage have become increasingly affordable enabling the discovery of hidden information from massive datasets.With ever growing databases it is important to find efficient data mining methods.This research concerns high-dimensional data that contains a high number of measured variables from the point of view of analysis methods, but also from the human perspective.Dimensionality reduction is a process where new features are extracted from high-dimensional data so that their number is smaller than in the input data while the information content stays the same.Diffusion map is a dimensionality reduction method suitable for data exhibiting nonlinear behavior.This research shows that diffusion map data mining technology can be brought to diverse application areas, e.g., to clustering and to system health monitoring.Several studies concerning these data mining tasks are presented.Firstly, clustering using diffusion maps is demonstrated with text mining and brain imaging data.These studies show that the methodology is suitable for discovering knowledge and finding structure in datasets coming from quite different sources.Secondly, system health monitoring using diffusion maps is presented in network security and mechanical engineering scenarios.The related studies show that abnormal behavior in the systems can be found with the proposed methodologies.
Author: Tuomo Sipola
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
Award: G6441683281
OpenAlex ID: W2236942143