Agricultural University of Athens
Αυτοματοποιημένα εργαλεία εξόρυξης βιο-γεωδεδομένων μέσω μεθόδων εξελικτικής κατάτμησης χρονοσειρών
Abstract
dc:descriptionAn important problem arising while analyzing large time series data sets, both in dimension and size, relates to the proper selection of a subset of the original features. Preprocessing the time series to obtain a representative meta-data set not only significantly reduces computational time, but also functions as a smoothing technique to weed out possible non systematic portions of the initial information, which may, in an extent, inhibit the analytical process. Conventional methods of time series data preprocessing, such as segmentation along the time axis for fast response, nonlinear normalization to emphasize significant information, averaging samples of the plant virus waves to suppress noise effects, reduction in the number of samples to realize a more compact network, include descriptive statistical methods such as re-sampling techniques or moving average procedures, both of which manipulate the initial information in a fixed width fashion. On the other hand, time series analysis plays an important role for phytopathology and virology, especially as regards to virus identification, which is made possible due to time series assessment.The design, development and implementation of an innovative method is described in this manuscript, aiming to overcome the limitations posed by the fixed width of the analytical tools. The algorithm allows for the production of effective secondary data, after having preprocessed the original time series information in an evolutionary fashion. Thus, it drastically reduces the size of the raw data table to more compact sets of cases and, at the same time, retaining all the crucial information of the initial time-series. This is achieved by the development of analytical tools of evolutionary adaptive width, propelled by Genetic Algorithms, Artificial Neural Networks and Support Vector Machines.The proposed methodology was applied for the solution of two problems. In the first case, essentially a classification problem, the Bioelectric Recognition Assay method was engaged so as to provide information used in the detection and identification of certain plant viruses, namely the Tobacco Rattle Virus and the Cucumber Green Mottle Mosaic Virus, using appropriately preprocessed reagents as the sensing elements. While reacting to the biosensors, each of the viruses in question exhibit unique patterns of biosensor responses over specific ranges of concentrations, rendering these responses as a special characteristic for each virus, a real identification signature. Each signature is in essence a graphical curve of bioelectrical responses in the time unit, a time series data set, which should be identified as a characteristic for each virus and effectively classified.The second problem on which the method was applied relates to the management of water reservoirs. The island of Cyprus was elected as the study area, while the inputs of the problem include structural and dynamic data, in which monthly precipitation particles play a distinct role. In this case, the time series information originated from the historical monthly rainfall data measured at certain watershed stations for a wide temporal period. The issue here was to develop a methodology for the production of evolutionary training/testing data, in order to achieve an effective estimation of the Average Annual Water Supply index on an annual basis, for each mountainous watershed of Cyprus.
Degree
thesis:*- Grantor dc:publisher
- Agricultural University of Athens
- Year dc:date
- 2012
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Glezakos, Thomas
- Γκλεζάκος, Θωμάς
Subjects
dc:subject × 20- Εξόρυξη δεδομένων
- Νευρωνικά δίκτυα
- Μηχανές διανύσματος υποστήριξης
- Γενετικοί αλγόριθμοι
- Αναγνώριση φυτικών ιών
- Πρόβλεψη χειμαρρικών κινδύνων
- Επεξεργασία σήματος
- Εξαγωγή χαρακτηριστικών χρονοσειρών
- Data mining
- Neural networks
- Support vector machines
- Genetic algorithms
- Virus identification
- Torrential risk forecasting
- Signal processing
- Time series feature extraction
- Φυσικές Επιστήμες
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Natural Sciences
- Computer and Information Sciences
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/28472