Omar E. Melikechi: Integrated path stability selection
Abstract
Feature selection can greatly improve performance and interpretability in machine learning problems. For example, it has been used to identify genes that are associated with certain diseases. Stability selection is a popular method for improving feature selection algorithms. However, it often selects few features, resulting in a low true positive rate. In this talk, I will introduce a novel approach to stability selection, called integrated path stability selection (IPSS), that yields significantly more true positives in practice, while still controlling the number of false positives. Furthermore, IPSS is fast, effective in high dimensions, and easy to implement, requiring just one user-specified parameter: the target number of false positives or the target false discovery rate. After introducing the method, I will demonstrate its performance on cancer data.