03583ntm a22005177i 4500 000719738 CZ-PrVSE 20250301181534.0 m d cr n|||||||||| 250301s2024 xr fsbm 000 0 eng d NEZPRACOVANÝ IMPORT ABA006 cze ABA006 ABA006 rda Fedorov, Mikhail ISIS:162172 dis Porovnání frekventistické a Bayesovské logistické regrese na příkladu reálných dat eng Comparison between Frequentist and Bayesian logistic regression on the example of real data / Mikhail Fedorov 2024 ?? stran : digital, PDF soubor Vedoucí práce: Tomáš Karel Bakalářská práce (Bc.)—Vysoká škola ekonomická v Praze. Fakulta informatiky a statistiky, 2025 Obsahuje bibliografii Textový (vysokoškolská kvalifikační práce) Rok obhajoby 2025 This study compares frequentist and Bayesian approaches to logistic regression using real-world data, with a particular focus on the impact of informative and non-informative priors in Bayesian analysis. The research is motivated by advances in computational capabilities that have made Bayesian methods increasingly accessible in machine learning applications. The methodology involved splitting the dataset into two parts: a primary subset (20%) for model evaluation and a larger auxiliary subset (80%) simulating prior knowledge. Three approaches were compared: classical logistic regression, Bayesian logistic regression with non-informative priors, and Bayesian logistic regression with informative priors. Model performance was evaluated using multiple metrics including Accuracy, Precision, Recall, F1-score, and ROC-AUC score, along with an analysis of 95% confidence intervals (CI) as well as credible intervals . Results demonstrate that while classical and non-informative prior Bayesian approaches showed similar performance, the Bayesian model with informative priors achieved superior results with an accuracy of 0.83 (compared to 0.78), F1-score of 0.85 (versus 0.81), and notably higher ROC-AUC of 0.93 (versus 0.81). The most significant improvement was observed in parameter uncertainty estimation, where the credible interval length undergone an average reduction of 15.1% compared to the classical approach, with improvements ranging from 10.3% to 28.2% across different variables. Způsob přístupu: Internet datové analýzy a modelování [obor bakal. práce] bakalářské práce fd132403 czenas bachelor's theses eczenas Bayesian statistics Logistic regression Informative priors Machine learning Confidence intervals Karel, Tomáš vse2015858033 ths Bílková, Diana mzk2004248684 opn Vysoká škola ekonomická v Praze. Fakulta informatiky a statistiky kn20010709399 dgg https://insis.vse.cz/zp/88781/podrobnosti VŠKP v InSIS https://insis.vse.cz/zp/88781 Hlavní práce https://insis.vse.cz/zp/88781/posudek/vedouci Hodnocení vedoucího https://insis.vse.cz/zp/88781/posudek/oponent/84888 Oponentura https://insis.vse.cz/zp/88781/podrobnosti dc:identifier NEPOSILAT VSKP vse88781 250205 88781