Full text
| |
| Author(s): |
Adriano Arantes Paterlini
Total Authors: 1
|
| Document type: | Master's Dissertation |
| Press: | São Carlos. |
| Institution: | Universidade de São Paulo (USP). Instituto de Ciências Matemáticas e de Computação (ICMC/SB) |
| Defense date: | 2011-03-28 |
| Examining board members: |
Caetano Traina Junior;
Renato Fileto;
Mario Antonio do Nascimento
|
| Advisor: | Caetano Traina Junior |
| Abstract | |
The success of Database Management System (DBMS) for applications with traditional data (numbers and short texts) has encouraged its use in new types of applications that require manipulation of complex data. Time series, scientific data and other multimedia data are examples of complex data. Several application fields, like medical informatics, have demanded solutions for managing complex data. Complex data can also be studied by means of Knowledge Discovery Techniques (KDD) applying appropriate clustering algorithms. However, these algorithms have high computational cost hindering their use in large data sets. The techniques already developed in the Databases research field for indexing metric spaces usually consider the sets have a uniform distribution, without taking into account the existence of clusters in the data, therefore the structures need to generalize the efficiency of queries for the entire set simultaneously. However the similarity searching is often limited to a specific region of the data set. In this context, this dissertation proposes a new access method able to index metric data efficiently, especially for sets containing clusters. It also proposes a new algorithm for clustering metric data so that selection of a medoid from a particular subset of elements becomes more efficient. The experimental results showed that the proposed algorithms FAMES and M-FAMES can be used as a clustering technique for complex data that outperform PAM, CLARA and CLARANS in effectiveness and efficiency. Moreover, the similarity searching performed with the proposed metric access method FAMESMAM proved to be especially appropriate to data sets with clusters (AU) | |