Tehnike učenja za klasifikaciju bioloških nizova

master's thesis

• Author: Marija Pezić
• Mentor: Pavle Goldstein (mentor)
• Committee member: Pavle Goldstein (predsjednik povjerenstva)
• Committee member: Juraj Šiftar (član povjerenstva)
• Committee member: Zrinka Franušić (član povjerenstva)
• Committee member: Rudi Mrazović (član povjerenstva)
• Granter: University of Zagreb; Faculty of Science; (Department of Mathematics); Zagreb
• Defense date and country: 2020-02-27, Croatia
• Scientific / art field, discipline and subdiscipline: NATURAL SCIENCES; Mathematics

Abstract

Cilj rada je analiza linearne klasifikacije proteina pomoću max-norme i algoritma stroja potpornih vektora. Na početku rada dani su osnovni matematički pojmovi iz linearne algebre, optimizacije, vjerojatnosti i statistike potrebni za razumijevanje rada. Također, opisan je rad algoritma stroja potpornih vektora. Za analizu se koristi sedam proteinskih familija. Skup za učenje sačinjen je od po 300 proteina iz svake proteinske familije, dok je skup za testiranje sačinjen od po 100 proteina iz svake proteinske familije. Objašnjen je postupak određivanja značajki svake familije te način izrade modela koji opisuju odnos svakog proteina sa zadanim proteinskim familijama. U konačnici su provedeni testovi linearne klasifikacije pomoću max-norme odnosno algoritma stroja potpornih vektora za različite parametre, te je napravljena usporedba dobivenih rezultata. Pokazalo se da su rezultati klasifikacije za različite metode gotovo isti i jako dobri.

Abstract (english)

The aim of this work is an analysis of protein classification using the max-norm and support vector machine algorithm. First, we present some basic concepts from linear algebra, optimization, probability, and statistics that are needed for understanding later. Also, we explained mathematical background of the support vector machine algorithm. There are 7 protein families used for analysis. The training set consists of 300 proteins from each family, and the test set consists of 100 proteins from each family. We described the process of determining features for each family and how to build a model that describes the relation between given protein and protein families. Finally, we implemented the linear classification using the max-norm and support vector machine algorithm for different parameter values. It is shown that results are almost the same for both algorithms, and classification is very accurate.

• Keywords: linearna klasifikacija proteina; algoritam stroja potpornih vektora; max norma
• Keywords (english): protein classification; support vector machine algorithm; max-norm
• Language: croatian
• URN:NBN: urn:nbn:hr:217:005111
• Study programme: Title: Mathematical Statistics; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra matematike (magistar/magistra matematike)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2020-09-23 10:57:58