Feature Selection via Sparse Regression for Classification of Functional Brain Networks

Despite the ongoing progress to chart the differences between the healthy controls and patients at the group level, the pattern classification of functional brain networks across individuals is still a challenging task. The difficulties include the very high dimensional feature space and very small sample size, as well as the probably high noise level. In this paper, we apply the stable sparse regression to pick the very few most discriminant features (edges) for the following classification. We considered different noise to signal ratios and sparsity controlling parameters and numerical experiments based on simulated data demonstrate the much better classification performance via the feature selection based on the sparse regression.