C. Lancho Martín
Complexity measures are aimed at characterizing the underlying complexity of supervised data. These measures provide insights into the factors that can hinder the performance of classifiers, such as overlap, linearity or density.
Traditionally, complexity measures have been designed to estimate the complexity of the entire dataset. However, recent research has shifted towards developing or adapting measures that offer a more granular perspective of data complexity adding the instance or the class level. Following this path, the authors have proposed the hostility measure: a multi-level complexity measure that provides insights into the complexity of data at the instance, class, and dataset levels. Its Python implementation can be found at https://github.com/URJCDSLab/Hostility_measure.
Some R and Python packages implementing complexity measures at different levels are: the ‘EcoL’ and ‘ImbCoL’ R packages, the ‘problexity’ and the ‘PyHard’ Python packages.
Keywords: complexity measures, supervised classification problems, software implementation
Scheduled
GT18.SOFTW1 Invited Session
November 7, 2023 4:50 PM
HC1: Canónigos Room 1