Библиографическое описание:Modified Correlation-Based Feature Selection for Intelligence Estimation Based on Resting State EEG Data : доклад, тезисы доклада / T. Avdeenko, A. Timofeeva, M. Murtazina. - [S. l. : s. n.], 2022. - Текст : непосредственный // Advances in Swarm Intelligence : 13th International Conference, ICSI 2022, Proceedings, Part II. - Zug, 2022. - 13345, Part II. - P. 289-300. - The research is supported by Ministry of Science and Higher Education of Russian Federation (project No. FSUN-2020–0009). - ISBN 9783031097263, DOI 10.1007/978-3-031-09726-3_26.
Аннотация:The effectiveness of the correlation-based method (CFS) for feature selection based on electroencephalogram (EEG) data of the resting state for the purpose of intelligence assessment is investigated. A modification of the CFS is proposed, which makes it possible to vary the cardinality of a subset of selected features using a hyperparameter. A practical example of the analysis of the relationship between the intelligence quotient (IQ), the age of subjects, the features extracted from EEG data, and the effects of their interaction is considered. A comparison of the genetic algorithm and the forward selection was made to find the optimal subset of features within the modified CFS. It was found that it is quite sufficient to use the method of forward selection. Using the nested cross-validation procedure, it was shown that the modified approach gives a lower mean absolute error compared to usual CFS, as well as building a stepwise regression by the forward selection method based on the Bayesian information criterion (BIC). In terms of the mean absolute error, the modified CFS is close to the least absolute shrinkage and selection operator (LASSO) and the improved algorithm Bolasso-S.
