20.500.12556/DKUM-70600 Sistem za avtomatizirano ocenjevanje časovnih okvirjev implementacije funkcionalnosti pri pripravi ponudb A system for automated evaluation of time estimates for implementation of functionalities in the process of offer preparation V magistrski nalogi predstavljamo spletno aplikacijo, ki s pomočjo metod strojnega učenja napoveduje predvidene časovne okvirje posameznih funkcionalnosti pri pripravi ponudb za stranke podjetja. Aplikacija rešuje problem z regresijo in za napovedovanje uporablja tri različne algoritme, in sicer linearno regresijo, k-najbližjih sosedov in metodo podpornih vektorjev. Algoritmi modele naučijo na osnovi že dokončanih časovnih okvirjev. Ob vsakem novem podatku, ki ustreza pogojem, da ga uvrstimo v učno množico, se modeli na novo naučijo. Algoritme smo med seboj primerjali s tremi merami uspešnosti. To so koren srednje kvadratne napake, srednja absolutna napaka in korelacijski koeficient. Raziskovali smo hipotezo, da lahko s strojnim učenjem napovemo podobne ocene, kot jih je podal človek. Izmed izbranih algoritmov je najbolj natančne rezultate podala metoda podpornih vektorjev. S primerjavo mer uspešnosti in odstopanj med napovedmi modelov algoritmov in človeka smo prišli do zaključka, da lahko našo hipotezo potrdimo. Rezultati so pokazali, da so modeli algoritmov podali dovolj natančne ocene, saj napake pri napovedi ne bi imele neposrednega vpliva na izvedbo projekta. In the master's thesis we present an online application that, with the help of methods of machine learning, predicts the anticipated time estimates of individual functionalities in offer preparations for the clients of the company. The application solves the regression problem and uses three different algorithms for predicting: linear regression, k-nearest neighbours, and support vector machine. Algorithms teach models on the basis of already completed time estimates. Each new data, which meets the conditions for classifying it into a learning set, the models re-learn. The algorithms were compared with three performance metrics. These are a root mean square error, a mean absolute error and a correlation coefficient. We have explored the hypothesis that with machine learning, we can predict similar estimates as man has done. Among the selected algorithms, the most accurate results were given by the support vector machine. By comparing the merit of success and deviations between forecasts of algorithms and man, we have come to the conclusion that our hypothesis can be confirmed. The results showed that the models of algorithms gave sufficiently precise estimates, as the errors in the forecast would have no direct impact on the implementation of the project. priprava ponudb ocenjevanje časovnih okvirjev strojno učenje regresija offer preparation evaluation of time estimates machine learning regression true true false R. Novak Slovenski jezik Angleški jezik Magistrsko delo/naloga 2018-05-29 20:53:12 2018-06-26 13:47:47 2022-08-01 19:33:19 2019-07-11 18:13:00 2018 0 [Maribor 0 0000-00-00 NiDoloceno NiDoloceno NiDoloceno 0000-00-00 0000-00-00 0000-00-00 004.58:004.728.8(043.2) 21512214 URN:SI:UM:DK:2C9PMCWS MAG_Novak_Rok_2018.pdf MAG_Novak_Rok_2018.pdf 1 CC0925FBB7488A859BE5C4D22226C95A 1467f2813c14040d2e81708a94da626e823cc22ba6c889ad76460588487a153e fbdc2e47-7c0f-11eb-bb7a-00155d0001ca 20.500.12556/dkum/3dd9a19b-ff9d-49ae-94e4-e6810bbada85 https://dk.um.si/Dokument.php?lang=slv&id=125113 Fakulteta za elektrotehniko, računalništvo in informatiko 0 0 0