A framework for multi-label exploratory data analysis: ML-EDA
Victor Augusto Moraes Carvalho, Newton Spolaôr, Everton Alvares Cherman, Maria Carolina Monard · 2014
Most supervised learning methods consider that each dataset instance is associated with a unique label. However, there are several domains in which the instances are associated with a set of labels (a multi-label). An alternative to investigate properties of multi-label data and their relationship with the learning performance consists in exploratory data analysis. This approach aims to obtain a better understanding of the data by using different techniques, most of them related to graphic representations. This work proposes ML-EDA, a framework for multi-label exploratory data analysis, which is publicly available in the Internet. The framework has been designed considering extensibility and maintainability as its main goals. Moreover, ML-EDA can directly process, among others, the information provided by MULAN, a framework for multi-label learning frequently used by the community. Some of the ML-EDA facilities are illustrated using benchmark multi-label datasets, highlighting its use as an additional resource to investigate multi-label data.