Aristotle University Of Thessaloniki (AUTH)
Ανακάλυψη γνώσης από ακολουθίες και δεδομένα συναλλαγών
Abstract
dc:descriptionAlthough “Knowledge Discovery in Databases” is a young field of research and application, having its roots in the late 1980s, it is already mature enough to offer useful analytical tools, embedded in popular software packages, for all kinds of busi-nesses and organizations. Its applications cover the entire spectrum of human activity, from healthcare and molecular biology to banking, commerce, industry and educa-tion. This growing variety of applications along with the vast amounts of storage and computing power available increases the need for more algorithms that are applica-tion-specific, faster and more effective. The importance of Knowledge Discovery in Databases (KDD) is further established by the fact that the progress and development of several fields (such as Molecular Biology) largely depends on its ability to deliver reliable and powerful techniques. Considering also that general purpose algorithms often fail to reach the performance standards set by real-world applications and the fact that KDD is a semi-automatic procedure that requires human guidance and inter-vention, one can understand the urgent need for new, specialized techniques as well as the improvement of existing ones. This thesis concerns a set of techniques for knowledge discovery from sequen-tial and transactional data. Both data categories share a common property, namely or-der, along (at least) one dimension. Specifically, two chapters of this thesis refer to the first data category. In the first one, a novel algorithm is proposed for the discov-ery of weak periodicities in discrete time series data. The algorithm is based on the circular autocorrelation function for the extraction of a refined set of candidate peri-odicities, used for mining periodic patterns. The second chapter that concerns sequen-tial data belongs to the research field of Bioinformatics. A new data mining metho-dology is developed for the prediction of the initiation site of the biological process of translation, in genome sequences. The proposed methodology identifies the main components of this problem and assigns a computational component to each separate-ly, in a modular fashion, achieving a high level of accuracy. The contribution of the thesis to knowledge discovery from transactional data is also twofold, covering two chapters. In the first one, a new algorithm is proposed for the prediction of rare events. The algorithm adopts an intuitive framework, em-beds the temporal information of the transaction in its items, delivering predictive patterns timely and effectively. In the second chapter of this category, the novel prob-lem of mining for mutually exclusive items is introduced and formulated and a set of metrics, along with a mining algorithm are provided.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2007
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Μπερμπερίδης, Χρήστος
- Berberidis, Christos
Subjects
dc:subject × 16Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/15264