Screening rules and information criteria-based analysis of gene expression data

High-dimensional data is becoming increasingly common, and the biomedical field is no exception with the rapid development of technology. There are various methods to deal with high-dimensional gene expression data, but all of them have some shortcomings. In this paper, we address the theory and application of feature screening in ultra-high-dimensional discriminative classification data, with the aim of reducing ultra-high-dimensional data to a size appropriate for the originally proposed sample size, while retaining all important variables. To this end, we propose a variable screening method that sure independence screening methods in conjunction with EBIC information criteria, which can effectively reduce data dimensionality while improving computational efficiency and helping to discover the most informative variables relevant to the target. In this paper, a random simulation sampling method was first used to select parameters and filter variables using randomly sampled data, and the correct selection rate and correct fit rate of the simulation results were higher than those of other approaches, which verified the reliability of the method used. Finally, four sets of real gene expression data were used to further validate the effectiveness of the method in selecting gene expression data features.