Stability selection for LASSO with weights based on AUC
Yonghan Kwon, Kyunghwa Han, Young Joo Suh, Inkyung Jung
Abstract
Stability selection is a variable selection algorithm based on resampling a dataset. Based on stability selection, we propose weighted stability selection to select variables by weighing them using the area under the receiver operating characteristic curve (AUC) from additional modelling. Through an extensive simulation study, we evaluated the performance of the proposed method in terms of the true positive rate (TPR), positive predictive value (PPV), and stability of variable selection. We also assessed the predictive ability of the method using a validation set. The proposed method performed similarly to stability selection in terms of the TPR, PPV, and stability. The AUC of the model fitted on the validation set with the selected variables of the proposed method was consistently higher in specific scenarios. Moreover, when applied to radiomics and speech signal datasets, the proposed method had a higher AUC with fewer variables selected. A major advantage of the proposed method is that it enables researchers to select variables intuitively using relatively simple parameter settings.