03797ntm a22005177i 4500 000553215 CZ-PrVSE 20190203085345.0 m d cr n|||||||||| 190203s2018 xr fsbm 000 0 eng d NEZPRACOVANÝ IMPORT ABA006 cze ABA006 ABA006 rda Panýr, Jakub ISIS:99692 dis Are judges good at estimating the future? : An application of machine learning on custody decision eng Are judges good at estimating the future? : An application of machine learning on custody decision / Jakub Panýr 2018 ?? stran : digital, PDF soubor Vedoucí práce: Jan Vávra Diplomová práce (Ing.)—Vysoká škola ekonomická v Praze. Národohospodářská fakulta, 2019 Obsahuje bibliografii Textový (vysokoškolská kvalifikační práce) Rok obhajoby 2019 This thesis aims to analyze and possibly improve the custodial decision in the CzechRepublic, as this kind of economic decision leads to social costs resulting from crime andfrom incarceration and because of that an improvement would be beneficial.Analysis itself is performed using machine learning techniques on a large dataset of custodialdecisions from 2008–2017. Gradient boosted decision tree algorithm is used becauseof its predictive power, combined with so-called contraction procedure, that serves as a solutionto missing labels problem, i.e. to the fact that I do not know whether people would havecommitted crime had they been released.This procedure is summarized by answering a simple question if the original situationcan be improved starting with the set of released people by the most benevolent group ofjudges and detaining additional people according to their predicted probability. Methodologyof Kleinberg et al. (2018) is closely followed and adapted to Czech Context.It turns that it is possible to improve the original situation, as main model was able todecrease crime rates by roughly three to four percents with no change in jailing rates. Thiswas done using the model with a quasi-random assignment of cases to judges. Besides that,this procedure allows me to solve problem with the possibly various preferences of judges asI am able to formulate model irrespectively of these preferences.One of the possible explanations of the fact that there is a potential to improve custodialdecisions (hence the explanation of the fact judges are somehow mistaken) stems from thefact that judges may be too dependent on a noise such as existence of unobservables and soon. However this complex problematic needs more research. Způsob přístupu: Internet ekonomická analýza [obor dipl. práce] diplomové práce fd132022 czenas master's theses eczenas economics of crime custody decision machine learning gradient boosting contraction procedure Vávra, Jan ISIS:85823 ths Kalíšková, Klára jo2013797871 opn Vysoká škola ekonomická v Praze. Národohospodářská fakulta kn20010709401 dgg https://insis.vse.cz/zp/66046/podrobnosti VŠKP v InSIS https://insis.vse.cz/zp/66046 Hlavní práce https://insis.vse.cz/zp/66046/posudek/vedouci Hodnocení vedoucího https://insis.vse.cz/zp/66046/posudek/oponent/59777 Oponentura https://insis.vse.cz/zp/66046/podrobnosti dc:identifier NEPOSILAT VSKP vse66046 190202 66046