Simulation Study of Imbalanced Classification on High-Dimensional Gene Expression Data

Purpose: Classification of gene expression helps study disease. However, it faces two obstacles: an imbalanced class and a high dimension. The motivation this is to examine the effectiveness undersampling before feature selection on high-dimensional data with classes.Methods: Least Absolute Shrinkage Selection Operator (Lasso), which can select features, handle modeling. Random (RUS) be used deal classes. Decision Tree (CART) algorithm construct classification model because produce interpretable model. Thirty simulated datasets varying imbalance ratios are test proposed approaches, Lasso-CART RUS-Lasso-CART. generated from parameters real data.Results: simulation results show that when minority accounts for more than 25% observation size, method appropriate. Meanwhile, RUS-Lasso-CART effective size at least 20 observations.Novelty: novelty using hybrid address problem