Advanced feature selection methods for high dimensional data

Feature selection (FS) has become a significant part of the data processing pipeline.FS techniques reduce the original feature space without transformation, so that the original features are preserved and cogent interpretation is possible

Funding agency: The Ministry of Education, Science, Research and Sport of the Slovak Republic VEGA 1/0327/20

Principal investigator:doc. Ing. Peter Drotár, PhD.

Funding: 30 000 EUR

Project duration: 2020 - 2022

Status: Ongoing

Increasing dimensionality of data is tightly related to the phenomenon known as curse of dimensionality. Curse of dimensionality negatively impacts data mining algorithms. To avoid negative degradation of performance data preprocessing techniques such as feature selection are applied. The primary goal of the grant proposal is to develop novel feature selection methods, utilizing concepts such as transfer learning or ensembles, for high-dimensional data with focus on the most challenging domain: high dimensional -small sample data. Big size of the data gave rise to a new phenomenon in data processing: deep learning. However, there are many domains where only small number of samples are available and deep learning is not feasible. To overcome this issue the transfer learning is adopted recently as the very promising solution. Therefore, we will focus on transfer learning in context of deep learning and develop the feature selection methods that would allow for more efficient deep transfer learning.