2014
EMSS 2014 Proceeding
Searching and indexing distorted data collections
Authors: Tomas Kocyan, Jan Martinovic, Michal Podhoranyi
Abstract
Success of many models and artificial intelligence methods strongly depends on ability to quickly and precisely search input data collection. Despite the existence of many algorithms for faster searching, the most of them fail while processing distorted input. Unfortunately, the distortion is natural for many types of data collections, especially for measurements of natural phenomena such as precipitations, river discharge volume etc. In this type of collections, there are no exact levels for generated values. This paper discusses possibilities of indexing and searching such distorted inputs and also proposes an alternative approach for their indexing. The proposed approach utilizes the Voting Experts algorithm for splitting the input regarding statistical indicators, the Dynamic Time Warping for dealing with distorted inaccuracies and hierarchical clustering for grouping similar sequences. Finally, the sample result of proposed algorithm applied on data collections consisting of measured river discharge volumes is shown.
I3M Scientific Sponsors
I3M Industrial Sponsors
I3M Media Sponsors
