|
| |
|
|
Robust regression and robust neural networks
Janáček, Patrik ; Kalina, Jan (advisor) ; Maciak, Matúš (referee)
The classical least squares approach in linear regression is prone to the presence of outliers in the data. The aim of this thesis is to present several robust alternatives to the least squares method in the linear regression framework and discuss their properties. Then robust neural networks based on these estimators are introduced and compared in a simulation study. In particular, the least weighted squares method with adaptive weights seems promising, as it is able to combine high robustness with efficiency in the absence of contamination in the data. 1
|
|
|
A Bootstrap Comparison of Robust Regression Estimators
Kalina, Jan ; Janáček, Patrik
The ordinary least squares estimator in linear regression is well known to be highly vulnerable to the presence of outliers in the data and available robust statistical estimators represent more preferable alternatives. It has been repeatedly recommended to use the least squares together with a robust estimator, where the latter is understood as a diagnostic tool for the former. In other words, only if the robust estimator yields a very different result, the user should investigate the dataset closer and search for explanations. For this purpose, a hypothesis test of equality of the means of two alternative linear regression estimators is proposed here based on nonparametric bootstrap. The performance of the test is presented on three real economic datasets with small samples. Robust estimates turn out not to be significantly different from non-robust estimates in the selected datasets. Still, robust estimation is beneficial in these datasets and the experiments illustrate one of possible ways of exploiting the bootstrap methodology in regression modeling. The bootstrap test could be easily extended to nonlinear regression models.
|
|
|
First order optimization methods in machine learning problems
Janáček, Patrik ; Branda, Martin (advisor) ; Kozmík, Karel (referee)
The goal of the thesis is to introduce the stochastic gradient method for optimizing differentiable objective function and discuss its convergence. First, supervised learning and empirical risk minimization (ERM) are explained. Then stochastic gradient descent (SG) is itroduced and analysed, first in the context of strictly convex objective function and then for the general non-convex function. In the last part, the classification of email spam is practically solved. 1
|
|
|
On the Effect of Human Resources on Tourist Infrastructure: New Ideas on Heteroscedastic Modeling Using Regression Quantiles
Kalina, Jan ; Janáček, Patrik
Tourism represents an important sector of the economy in many countries around the world. In this work, we are interested in the effect of the Human Resources and Labor Market pillar of the Travel and Tourism Competitiveness Index on tourist service infrastructure across 141 countries of the world. A regression analysis requires to handle heteroscedasticity in these data, which is not an uncommon situation in various available human capital studies. Our first task is focused on testing significance of individual variables in the model. It is illustrated here that significance tests are influenced by heteroscedasticity, which remains true also for tests for regression quantiles or robust regression estimators, resistant to a possible contamination of data by outliers. Only if a suitable model is considered, which takes heteroscedasticity into account, the effect of the Human Resources and Labor Market pillar turns out to be significant. Further, we propose and present a new diagnostic tool denoted as aquintile plot, allowing to interpret immediately the heteroscedastic structure of the linear regression model for possibly contaminated data.
|
guest ::
RSS feed.