OBERTA, D. Statistické modely pro predikci časové náročnosti projektů [online]. Brno: Vysoké učení technické v Brně. Fakulta strojního inženýrství. 2023.

Posudek vedoucího

Hübnerová, Zuzana

In his bachelor's thesis, Mr. Oberta focuses on the practical application of statistical regression methods on real data obtained from his collaboration with Hitachi Energy CZ, s.r.o. He evaluates the suitability of the linear and generalized linear models with regression trees, and subsequently random forests in modeling of project duration depening on various parameters of the project. From my perspective, it is admirable how deeply and extensively Mr. Oberta immersed himself in the knowledge of the utilized models in just a few months. The attention in the first part of the thesis is focused on the properties of estimations and the assessment of model suitability in the case of linear and generalized linear models. In the chapter on regression trees, Mr. Oberta pays more attention than the source literature to the precise description of the procedure for splitting the range of possible values of covariates, including categorical covariates. In the same chapter, you can also find a deeper clarification of the proof of the theorem on an optimally pruned tree (Theorem 4.10). The contribution of the thesis lies in the part dedicated to regression trees assuming gamma distribution of the explained variable. Unlike the commonly used residual sum of squares, a specific form of the deviance criterion used in maximum likelihood theory is derived here. Furthermore, the thesis presents the procedure for analysis in random forests, which utilize bootstrapping in the application of regression trees. Each chapter is complemented by the derivation of at least an asymptotic confidence interval for the mean of the explained variable, or a prediction interval (for the linear model). In the case of regression trees, the interval estimate of the mean of the explained variable is derived separately using the Student's distribution under the assumption of a fixed distribution corresponding to the assumed model. In the chapter Real Data Example, you can read the outputs of the analysis and the assessment of the suitability of individual models. For this purpose, it was necessary to program all the approaches, this time in Python. Yes, some functions are directly available in Python, however Mr. Oberta programmed regression trees and random forests, for example, from scratch to ensure that the program corresponds to the structure of his data or the assumptions placed on it. I have joyfully observed Mr. Oberta working independently and diligently on the chosen topic. It was an honor for me to supervise him. I'm also glad to add that Mr. Oberta participated with this thesis in the Competition of university students (SVOČ 2023) in the section Theory of probability, mathematical statistics, econometry, and financial mathematics.

Posudek oponenta

Žák, Libor

Bakalářská práce pana Oberty se zabývá hledáním statistických modelů pro predikci časové náročnosti projektů. Bakalářská práce se skládá ze dvou hlavních částí. V první části je teoretický popis metod lineární a zobecněné lineární modely. Jsou zde uvedeny základní vlastnosti modelů včetně predikčních intervalů. Tyto vlastnosti jsou uvedeny včetně důkazů, z nichž některé jsou přínosem autora. Dále je v první části diskutovány metody pro hledání vhodného regresního modelu. V druhé části aplikuje popsané metody na data popisující časovou náročnost projektů. Zabývá se porovnáním jednotlivých metod a rozebírá jejich přínosy (nedostatky). Práce je napsaná velmi kvalitně bez překlepů a chyb (snad jen obrázky s výsledky by mohly být lépe čitelné) a přesahuje požadavky kladené na BP. Bakalářskou práci pana Oberty doporučuji k obhajobě a hodnotím stupněm A.

eVSKP id 149614