Comparison of gene selection performances of boruta and elastic net algorithms on rna sequencing data: A monte carlo simulation study

Boruta Algorithm and Determan's Algorithm in Elastic Net Generalized Linear Models were applied to select the genes that have a significant effect on the related cancer disease using eight real RNA sequencing data set belonging to different cancer types obtained from the Cancer Genome Atlas Program. Algorithms belonging to seven of the methods within the Boruta Algorithm, which differ according to different classification methods and various importance criteria, were applied separately and the results obtained were compared not only with the Elastic Network Algorithm but also with each other. A Monte Carlo simulation study was performed based on the real data sets. The gene sets used for feature selection were obtained by applying preprocessing steps in four stages: filtering, normalization, transformation and univariate analysis. The performances of the feature selection methods were compared using measures such as Precision, Recall and F1 Measure. For all of the data sets used in the study, Elastic Net Algorithm stood out in terms of Precision. Boruta Algorithm based on Extra Trees and Random Ferns outperformed the Elastic Net Algorithm in terms of Recall.