DeepTD: Deep Neural Network-based Feature Selection with Local False Discovery Rate Estimation

Introduction

DeepTD is a novel error-controlled feature selection method for DNNs, which explores the target-decoy competition (TDC) strategy to estimate the local false discovery rate (FDR) of selected features. The merits of the proposed method include: confidence assessment of individual selected features by local FDR estimation; multiple competition with decoy features to improve statistical power; better robustness to small numbers of important features and low FDR thresholds than FDR control methods, e.g., knockoff-based methods; a new DNN-derived measure of feature importance.

Download

The source code for applying DeepTD can be downloaded here.

 

Contact Us

Any problem with DeepTD please contact:

Yan Fu: yfu@amss.ac.cn
Academy of Mathematics and Systems Science, Chinese Academy of Sciences, Beijing 100190, China.