Induktivno učenje iz opazovanj

diplomsko delo

Miha Pišorn (Author), Nikola Guid (Mentor), Damjan Strnad (Co-mentor)

Abstract

V diplomskem delu predstavimo učenje iz podatkov, kot model predvidevanja uporabimo odločitvena drevesa. Preučimo problem prekomernega prilagajanja in pogoste metode za njegovo omiljenje. Ansambelsko učenje je koncept v okviru umetne inteligence, ki združuje metode, ki sestavijo nabor klasifikatorjev in klasificirajo nove vhodne podatke na podlagi glasovanja. Te metode preučimo in pokažemo, zakaj se pogosto odrežejo bolje od posameznih klasifikatorjev. Implementiramo pogosto uporabljan algoritem Adaboost in preizkusimo njegovo obnašanje. Kot klasifikatorje uporabimo odločitvena drevesa.

Keywords

umetna inteligenca;strojno učenje;odločitvena drevesa;ansabelsko učenje;Adaboost;

Data

Language:	Slovenian
Year of publishing:	2014
Typology:	2.11 - Undergraduate Thesis
Organization:	UM FERI - Faculty of Electrical Engineering and Computer Science
Publisher:	[M. Pišorn]
UDC:	004.89(043.2)
COBISS:	18546710
Views:	1878
Downloads:	108
Average score:	0 (0 votes)
Metadata:

Other data

Secondary language:	English
Secondary title:	INDUCTIVE LEARNING FROM OBSERVATION
Secondary abstract:	In this diploma thesis we review learning from data using decision trees as a prediction model. We study the problem of overfitting and review common methods used to contain it. Ensemble learning is a concept in artificial intelligence that encompasses methods constructing a set of classifiers and classify new input data by taking a vote of their predictions. We review these methods and show why they often outperform single classifiers. We implement commonly used Adaboost algorithm and test its behavior, using decision trees as classifiers.
Secondary keywords:	artificial intelligence;machine learning;decision trees;ensemble learning;Adaboost;
URN:	URN:SI:UM:
Type (COBISS):	Bachelor thesis/paper
Thesis comment:	Univ. v Mariboru, Fak. za elektrotehniko, računalništvo in informatiko, Računalništvo in informacijske tehnologije
Pages:	34 f.
ID:	8738923