The effect of noise and sample size in the performance of unsupervised feature selection using mixture models

Orientació: recerca

Departament del director de la tesi: LSI
Director de la tesi: Alfredo Vellido
Correu electrònic: avellido@lsi.upc.edu

Descripció breu:
For multivariate data sets, some features are better than others at revealing the underlying cluster structure of the observed sample. Existing statistical learning models for unsupervised feature selection are prone to suboptimal performance due to the existence of noise in the observed data and / or due to small sample sizes. The current Master project will attempt to quantify the robustness of these methods under those circumstances and will aim to provide effective solutions to overcome such limitations.