We are proud to announce the presentation of the paper “Survey of dataset meta-features” authored by Dr. Dijana Oreški and Dr. Nikola Kadoić at the upcoming conference International Scientific Conference on Economic and Social Development hosted by Universidade Lusofona do Porto.
The full paper will be published in conference proceedings after the conference, and also on this webpage.
Abstract. Machine learning is an active area of research and has seen the development of many algorithms. Machine learning practitioners at different levels frequently face a similar problem: What algorithm best suits their data? Recent meta-learning research automates this procedure using a meta-classifier to predict the best algorithm for a given dataset. Meta-features are used to describe the properties and characteristics of datasets and construct the feature space for meta-learning. In this paper, we performed a literature survey to recognize meta-features that should be taken into account when providing meta-learning research and identify directions for further research in this area.
Keywords: Meta-learning, meta-features, dataset characteristics, machine learning.