Full text
| |
| Author(s): |
Elisângela Botelho
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: | 2002-08-09 |
| Examining board members: |
Caetano Traina Junior;
Carla Maria Dal Sasso Freitas;
Rosely Sanches
|
| Advisor: | Caetano Traina Junior |
| Abstract | |
The automation of the activities in several areas, sueh as business, engineering, medicine, science and government, are increasing everyday, pressing, the volume of data stored in the databases to grow exponentially. It is interesting to make use of these data beyond the original objectives of the enterprise. Moreover, it is desirable to extract useful information not previously foreseen, aggregating value to the enterprise. Although the database management systems supply basic tools to data recovery and analysis through the standard transactions in large amounts of data, analyzing these data in numerical or textual format, especially in spaces of high dimensions, is an overwhelming job to human beings. On the other hand, human being has a great capacity to quickly absorb and understand information represented of graphical form. As classification is one of the most common analysis tasks, this work aims at the development of a new technique for the visual construction of classifiers, using this high human capacity to analyze data represented in graphical format as factor to guide the training of the classifier. This work extends the FastMapDB tool, that in its original version allowed only the visualization of data and it did not make possible the user to interfere in the visualization process, to allow not only this visualization of data, but also, the reverse identification of mapped objects and the delimitation of object regions in the visualization. This makes it possible to recover the data present in the database that the user consider interesting from the visualization, and providing to the tool visual resources for classification of the new objects according to rules that the user can define in the mapped space of visualization. The proposal of this work is new in regarding to apply visualization techniques as a factor to train the classifier. Existing tools are restricted to show results already discovered by the system. Our system, in contrast, takes advantages from the interpretation obtained from the user will through its interaction with the visualizations presented, and allows that subsequent data analysis tasks automatically use these results. (AU) | |