Network-based dimensionality reduction of high-dimensional, low-sample-size datasets

In the field of data science, there are a variety datasets that suffer from high-dimensional, low-sample-size (HDLSS) problem; however, only few dimensionality reduction methods exist applicable to address this type problem, and is no nonparametric solution date. The purpose work develop novel network-based (nonparametric) analysis (NDA) method, can be effectively applied HDLSS data. First, with NDA correlation graph variables specified. With modularity-based community detection set modules Then, linear combination weighted by their eigenvector centralities (EVCs), defined as LVs, determined. optional phase variable selection, low EVCs communality ignored. LVs indicators belonging specified using method. publicly available databases compared principal factoring (PFA) methods. results show it outperforms existing in terms interpretability. addition, application easier, since need specify number latent due its nature. • A method proposed perform reduction. finds (LVs) indicators. provides feature ignoring common communalities. both LVs. tested component on databases.