A0450
Title: Dealing with count zeros in preprocessing of probability density functions
Authors: Stanislav Skorna - Palacky University Olomouc (Czech Republic) [presenting]
Jitka Machalova - Palacky University (Czech Republic)
Karel Hron - Palacky University (Czech Republic)
Abstract: Probability density functions occur naturally as data in applications in many different fields, e.g. income distribution in econometrics, mortality densities in epidemiology or soil contamination in geochemistry. However, since densities are mostly unobserved in their continuous form and available only through discrete samples, their reliable approximation is crucial for employing modern methods of functional data analysis. Zero counts or missing values are often presented in the samples, which causes the main issue in the preprocessing, e.g. for using centered log-ratio transformation. The contribution opens a discussion about the issue and presents possible solutions using adaptations of imputation methods for (multivariate) compositional data and sparse functional data analysis illustrated on simulated and real data.
