Nadzorovano strojno učenje s poudarkom na globokih nevronskih mrežah : delo diplomskega seminarja

Korat, Sara (author), Orbanić, Alen (mentor)

Namen diplomske naloge je predstaviti osnove nadzorovanega strojnega učenja in ene izmed najbolj uporabljenih metod nadzorovanega strojnega učenja (linearna regresija, logistična regresija, metoda k najbližjih sosedov, naključni gozdovi, metoda podpornih vektorjev, nevronske mreže). V prvem delu diplomske naloge so opisane glavne ideje metod strojnega učenja, pri čemer je osnovna matematična ideja podana samo pri linearni regresiji. Za ostale metode je poudarek na intuitivni razlagi. Opisane metode so predstavljene na primeru napovedovanja dobrih in slabih komitentov glede na dane podatke. Modeli za metode so zgrajeni v programu Weka, ki omogoča vizualen pregled podatkov in rezultatov. Poleg glavnih rezultatov, kot so metrike pravilnosti metode, Weka izpiše še različne statistične kazalce, ki merijo učinkovitost. V drugem delu je opisana metoda nevronskih mrež, njena uporabnost in implementacija na primeru, ki napoveduje, ali je komitent dober ali slab (gre za večje število podatkov kot v prejšnjem primeru). Namesto v programu Weka je model zgrajen v programskem jeziku Python s knjižnicama TensorFlow in Keras, ki omogočata večjo svobodo glede izbire števila mrež, števila nevronov in ostalih parametrov.

• Language: Slovenian
• Keywords: nadzorovano strojno učenje, Weka, nevronske mreže, Python
• Work type: Final seminar paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FMF - Faculty of Mathematics and Physics
• Year: 2018
• PID: 20.500.12556/RUL-103329
• UDC: 004
• COBISS.SI-ID: 18437209
• Publication date in RUL: 16.09.2018

KORAT, Sara, 2018, Nadzorovano strojno učenje s poudarkom na globokih nevronskih mrežah : delo diplomskega seminarja [online]. Bachelor’s thesis. [Accessed 14 April 2025]. Retrieved from: https://repozitorij.uni-lj.si/IzpisGradiva.php?lang=eng&id=103329

Secondary language

• Language: English
• Title: Supervised machine learning with emphasis on deep neural networks

Abstract

The purpose of this thesis is to present the basics of supervised machine learning and some of the most used methods of supervised machine learning (linear regression, logistic regression, k nearest neighbors, random forests, support vector machine, neural networks). The first part of the thesis describes the main ideas of the selected machine learning methods, where the basic mathematical idea is given only for linear regression, whereas for other methods, the emphasis is on intuitive explanation. The described methods are presented for the case of predicting good and bad customers based on the given data. Models for methods are built in a program named Weka, which allows for a visual display of data and results. In addition to the main results, such as the accuracy of the method, Weka prints out various statistical indicators that measure its effectiveness. The second part describes the neural networks method, its usage, and the implementation for an example that predicts whether the customer is good or bad for a bank (on a larger dataset than the previous example). This model is built in Python instead of Weka, which provides greater freedom in choosing the number of layers, the number of neurons, and other parameters.

• Keywords: supervised learning, Weka, neural networks, Python