Unsupervised Adaptation for High-Dimensional with Limited-Sample Data Classification Using Variational Autoencoder

keywords: HDLSS problem, dimensionality reduction, unsupervised framework, variational autoencoder, deep learning
High-dimensional with limited-sample size (HDLSS) datasets exhibit two critical problems: (1) Due to the insufficiently small-sample size, there is a lack of enough samples to build classification models. Classification models with a limited-sample may lead to overfitting and produce erroneous or meaningless results. (2) The 'curse of dimensionality' phenomena is often an obstacle to the use of many methods for solving the high-dimensional with limited-sample size problem and reduces classification accuracy. This study proposes an unsupervised framework for high-dimensional limited-sample size data classification using dimension reduction based on variational autoencoder (VAE). First, the deep learning method variational autoencoder is applied to project high-dimensional data onto lower-dimensional space. Then, clustering is applied to the obtained latent-space of VAE to find the data groups and classify input data. The method is validated by comparing the clustering results with actual labels using purity, rand index, and normalized mutual information. Moreover, to evaluate the proposed model strength, we analyzed 14 datasets from the Arizona State University Digital Repository. Also, an empirical comparison of dimensionality reduction techniques shown to conclude their applicability in the high-dimensional with limited-sample size data settings. Experimental results demonstrate that variational autoencoder can achieve more accuracy than traditional dimensionality reduction techniques in high-dimensional with limited-sample-size data analysis.
mathematics subject classification 2000: 68-T99
reference: Vol. 40, 2021, No. 1, pp. 1–28