03670ntm a22005657i 4500 000722948 CZ-PrVSE 20250608110309.0 m d cr n|||||||||| 250608s2025 xr fsbm 000 0 eng d NEZPRACOVANÝ IMPORT ABA006 cze ABA006 ABA006 rda Bogdan, Adam ISIS:149266 dis Modeling lapse rate in life insurance : A study of data balancing methods eng Modeling lapse rate in life insurance : A study of data balancing methods Adam Bogdan 2025 ?? stran : digital, PDF soubor Vedoucí práce: Michal Černý Diplomová práce (Ing.)—Vysoká škola ekonomická v Praze. Fakulta informatiky a statistiky, 2025 Obsahuje bibliografii Textový (vysokoškolská kvalifikační práce) Rok obhajoby 2025 Lapse rates have generally been estimated on product to product base. Using this approach, however, may be imprecise as it doesn't reflect each customers probabilities and therefore may negatively impact the cash flow predictions. Using real data from a Czech insurance company, this paper is aimed at generating individual probabilities as opposed to the estimated lapses for the whole product. Models like logistic regression, XGB, lasso, ridge, elastic net and random forest are estimated, tested and compared. The inevitable problem of unbalanced data is solved by over-sampling and/or under-sampling methods and their effects on predictions is deeply studied, thus more than 70 balanced training datasets are created. The main finding are that based on the provided dataset the various classification methods perform at fairly similar level. Since the balancing methods are based on K-Nearest Neighbors, the varying values of K were tested and it can be concluded that their parametrization does not significantly improve models. Comparing the main over-sampling methods, ADASYN methods had a higher F1 score than SMOTE methods. Also using hybrid methods did not prove to significantly improve prediction results compared to the individual methods. Overall, even the best models, do not perform at a level that would be deemed useful in real world use, judging by F1 scores that were at best around 0,2. Způsob přístupu: Internet ekonometrie a operační výzkum [obor dipl. práce] diplomové práce fd132022 czenas master's theses eczenas lapses XGB predictions data imbalance Random Forest logistic regression Černý, Michal mzk2005275013 ths Dvořáčková, Lucie ISIS:118793 opn Vysoká škola ekonomická v Praze. Fakulta informatiky a statistiky kn20010709399 dgg https://insis.vse.cz/zp/88312/podrobnosti VŠKP v InSIS https://insis.vse.cz/zp/88312 Hlavní práce https://insis.vse.cz/zp/88312/posudek/vedouci Hodnocení vedoucího https://insis.vse.cz/zp/88312/posudek/oponent/86671 Oponentura https://insis.vse.cz/zp/88312/priloha/31138 Přiloha k práci https://insis.vse.cz/zp/88312/priloha/31139 Přiloha k práci https://insis.vse.cz/zp/88312/priloha/31140 Přiloha k práci https://insis.vse.cz/zp/88312/podrobnosti dc:identifier NEPOSILAT VSKP vse88312 250606 88312