Přístupnostní navigace
E-application
Search Search Close
Master's Thesis
Author of thesis: Bc. Robin Stloukal
Acad. year: 2025/2026
Supervisor: doc. Mgr. Zuzana Hübnerová, Ph.D.
Reviewer: Ing. Matej Benko
Generalized linear models (GLMs) accommodate non-normal response variables, yet constructing prediction intervals (PIs) within them remains underexplored. This thesis investigates and compares PI construction methods in Poisson and gamma GLMs, evaluating conditional empirical coverage and interval width through extensive simulations in R. The core contribution lies in exploiting special predictive distributions derived from the properties of GLM estimators and the inversion of probabilistic inequalities. For the intercept-only Poisson model, a Bayesian PI utilizing a log-normal posterior distribution is introduced, alongside interval estimators based on Chernoff bounds, Chebyshev's inequality and its Vysochanskij–Petunin refinement, and Harremoës' sharp tail bounds. Two resampling algorithms based on the bootstrap and jackknife are also proposed. These methods are extended to Poisson log-linear regression, where contributions include one-sided, two-sided, and second-order Taylor-refined Chernoff intervals and an adapted Nelson's method, and further to gamma regression, where the Vysochanskij–Petunin inequality is applied to the gamma–log-normal predictive mixture. Simulation results reveal that discrete Poisson models require boundary randomization and benefit from exact and Bayesian approaches in small samples, while continuous gamma regression favors the classical asymptotic normal method and full conformal prediction, due to the presence of the dispersion parameter.
generalized linear model, prediction interval, Poisson regression, gamma regression, predictive mixture, conformal prediction, probability inequality
Date of defence
08.06.2026
Result of the defence
Defended (thesis was successfully defended)
Grading
A
Process of defence
Student odprezentoval svoji práci. Vedoucí a oponent přednesli své posudky. Oponent položil studentovi otázky z posudku, na které student uspokojivě odpověděl. Komise položila doplňující otázku, na kterou student zareagoval.
Language of thesis
English
Faculty
Fakulta strojního inženýrství
Department
Institute of Mathematics
Study programme
Mathematical Engineering (N-MAI-P)
Composition of Committee
prof. RNDr. Zdeněk Pospíšil, Dr. (předseda) prof. Mgr. Pavel Řehák, Ph.D. (místopředseda) doc. Mgr. Zuzana Hübnerová, Ph.D. (člen) doc. Mgr. Zdeněk Opluštil, Ph.D. (člen) doc. Mgr. Jaroslav Hrdina, Ph.D. (člen)
Supervisor’s reportdoc. Mgr. Zuzana Hübnerová, Ph.D.
Grade proposed by supervisor: A
Reviewer’s reportIng. Matej Benko
Grade proposed by reviewer: A
Responsibility: Mgr. et Mgr. Hana Odstrčilová