Métricas para el apoyo de la exploración visual de componentes en modelos de minería de datos

Abstract

The exploration of a Data Mining (DM) model, through the use of appropriate visual representation techniques and integrated interaction mechanisms, present advantages for the analyst or data miner when attempting to understand a data model. Currently, there are new proposals for methodologies and visualization schemes to support DM processes, which integrate features that combine DM techniques and ad-hoc graphic artifacts in order to facilitate the analysis and exploration of models, through the use of visualization in the input (exploratory data analysis) of the DM process, then in the model generation process (visualization and exploration of the model and its internal components), and finally in the output of this process (pattern visualization). However, this points to a qualitative and often subjective analysis, which depends directly on the experience and expertise of the analyst or data miner. In order to be able to complement this qualitative analysis, it is necessary to incorporate functions with metrics in the visual scheme that allow to corroborate it quantitatively. This work is oriented in this direction, and describes the definition, adaptation and implementation of a set of metrics that allow to validate and complement the visual analysis of an DM model, by using distance and similarity metrics, applied to the components of the MD model. This work uses as a case study, an DM model generated through the Decision Tree (DT) technique, combined with the Kohonen maps technique or Self-Organizing Map (SOM), applied to the components or nodes of the DT. It is possible to check the validity of the proposed metrics from their application on a known data set from a previously defined DM task.