Integración de técnicas de visualización a algoritmos de agrupamiento basados en densidad
Fecha
2011-06-20
Autores
Ruíz Quintana, Pablo Alberto
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Central “Marta Abreu” de Las Villas
Resumen
La mayoría de los métodos de agrupamiento requieren la especificación a priori de varios parámetros que repercuten significativamente en la calidad de la solución. En muchos casos los usuarios no disponen de la información necesaria para predecir valores óptimos para estos parámetros. No se conocen mecanismos adecuados para seleccionar de forma interactiva y en tiempo real los parámetros necesarios para obtener agrupamientos basados en densidad más eficientes. Un exponente claro de lo anterior lo constituye el algoritmo de agrupamiento DBSCAN, el cual tiene una alta complejidad temporal. El objetivo de la investigación consiste en valorar la efectividad de integrar técnicas de visualización a algoritmos de agrupamiento basados en densidad, en especial al algoritmo DBSCAN, para mejorar su eficiencia, a partir de la interacción en tiempo real con el usuario. Los principales resultados obtenidos son: se identificaron las interacciones a establecer en el algoritmo DBSCAN, se creó un modelo de integración de las técnicas de visualización y este algoritmo, se desarrolló el software DAVIA que implementa el modelo de integración y se ilustró mediante la experimentación con varias colecciones de objetos que se obtienen agrupamientos con mejores valores de las medidas de calidad internas y externas y en menor tiempo cuando se integran las técnicas de visualización con el algoritmo de agrupamiento DBSCAN, respecto a lo resultados cuando se aplica el algoritmo DBSCAN original sin visualización.
Most of the clustering methods require the specification a priori of several parameters that rebound significantly in the quality of the solution. In many cases the users don't have the necessary information to predict good values for these parameters. Appropriate mechanisms are not known to select in an interactive way and in real time the necessary parameters to obtain density based clusters more efficiently. A clear exponent of the above-mentioned constitutes it the clustering algorithm DBSCAN, which has a high temporary complexity. The objective of the investigation consists on valuing the effectiveness of integrating visualization techniques to density based clustering algorithms, especially to the algorithm DBSCAN, to improve its efficiency, starting from the interaction in real time with the user. The main results obtained are: were identified the interactions to settle down in the algorithm DBSCAN, was created a model of integration of the visualization techniques and this algorithm, was developed the software DAVIA that it implements the integration pattern and it was illustrated by means of the experimentation with several collections of objects that clusters are obtained with better values of the internal and external measures of quality and in smaller time when they are integrated the visualization techniques with the cluster algorithm DBSCAN, regarding that been when the original DBSCAN algorithm is applied without visualization.
Most of the clustering methods require the specification a priori of several parameters that rebound significantly in the quality of the solution. In many cases the users don't have the necessary information to predict good values for these parameters. Appropriate mechanisms are not known to select in an interactive way and in real time the necessary parameters to obtain density based clusters more efficiently. A clear exponent of the above-mentioned constitutes it the clustering algorithm DBSCAN, which has a high temporary complexity. The objective of the investigation consists on valuing the effectiveness of integrating visualization techniques to density based clustering algorithms, especially to the algorithm DBSCAN, to improve its efficiency, starting from the interaction in real time with the user. The main results obtained are: were identified the interactions to settle down in the algorithm DBSCAN, was created a model of integration of the visualization techniques and this algorithm, was developed the software DAVIA that it implements the integration pattern and it was illustrated by means of the experimentation with several collections of objects that clusters are obtained with better values of the internal and external measures of quality and in smaller time when they are integrated the visualization techniques with the cluster algorithm DBSCAN, regarding that been when the original DBSCAN algorithm is applied without visualization.
Descripción
Palabras clave
Integración, Técnicas de Visualización, Algoritmos de Agrupamiento, Densidad, Algoritmo DBSCAN, Software DAVIA, Inteligencia Artificial, Computación Gráfica