Strojno učenje : vrednotenje v zavarovalništvu : magistrsko delo

Erker, Matevž (author), Košir, Tomaž (mentor), Todorovski, Ljupčo (comentor)

V magistrskem delu si bomo pogledali posplošeni linearni model in njegove predpostavke. Kot že samo ime pove, je posplošeni linearni model posplošitev linearnega modela. Najpomembnejša posplošitev je predpostavka, da slučajna spremenljivka ni nujno porazdeljena normalno, ampak spada v družino eksponentnih porazdelitvenih funkcij. V drugem delu magistrskega dela se bomo posvetili strojnemu učenju in primerjanju metod strojnega učenja s posplošenim linearnim modelom. Kot vemo, se nahajamo v dobi podatkov. Edina rešitev za procesiranje in iskanje smisla v ogromni količini podatkov, ki je na voljo, je strojno učenje in podatkovno rudarjenje. Znanstveniki pravijo, da nekatere metode strojnega učenja posnemajo odločanje posameznikov. Tema tega magistrskega dela je tako poskus menjave posplošenega linearnega modela z modeli strojnega učenja. Pogledali si bomo, kako zgradimo odločitvena drevesa in kakšne parametre imamo, kako so zgrajene umetne nevronske mreže in njihovo povezavo z biološkimi nevronskimi mrežami ter kako se pri strojnem učenju odločamo za najboljši model. Na koncu je podan tudi primer izračuna štirih modelov (odločitveno drevo, naključni gozd, nevronske mreže in kaskadni model) ter primerjava s posplošenim linearnim modelom v programskem jeziku R.

• Language: Slovenian
• Keywords: posplošeni linearni model, strojno učenje, rudarjenje podatkov, prečno preverjanje, odločitvena drevesa, naključni gozdovi, nevronske mreže
• Work type: Master's thesis/paper
• Typology: 2.09 - Master's Thesis
• Organization: FMF - Faculty of Mathematics and Physics
• Year: 2018
• PID: 20.500.12556/RUL-103569
• UDC: 519.2
• COBISS.SI-ID: 18443097
• Publication date in RUL: 20.09.2018

ERKER, Matevž, 2018, Strojno učenje : vrednotenje v zavarovalništvu : magistrsko delo [online]. Master’s thesis. [Accessed 26 March 2025]. Retrieved from: https://repozitorij.uni-lj.si/IzpisGradiva.php?lang=eng&id=103569

Secondary language

• Language: English
• Title: Machine learning : pricing in actuarial science

Abstract

In the master's thesis we will examine the Generalized Linear Model (GLM) and its assumptions. As the name already implies, the GLM is a generalization of the linear model. The most important generalization is the assumption that the random variable is not necessarily distributed normally but belongs to the family of exponential distribution functions. In the second part of the master's thesis, we will look at machine learning and compare the methods of machine learning with GLM. As we know, we are living in the era of data. The only solution for processing and making sense of the vast amount of available data, is machine learning and data mining. Scientists say that some methods of machine learning mimic the decision-making of humans. The theme of this master's thesis is thus an attempt to replace GLM with one of the machine learning methods. We will look at how we can build decision trees and what parameters we have, how artificial neural networks are built, how they are related to biological neural networks, and how we can choose the best model in machine learning. Finally, an example of the calculation of four models (decision tree, random forest, neural networks and cascade model) as well as a comparison of these models with GLM in programming language R is presented.

• Keywords: generalized linear model, Machine learning, Data mining, cross-validation, decision tree, random forest, neural netwoks