No public access
undergraduate thesis
LEARNING COMPREHENSIBLE TREES DISGARDING COMPLEX INSTANCES

Ivana Barić (2016)
University of Rijeka
Department of Informatics
Metadata
TitleUČENJE STABALA ODLUČIVANJA ZANEMARUJUĆI KOMPLEKSNE PRIMJERE
AuthorIvana Barić
Mentor(s)Sanda Martinčić-Ipšić (thesis advisor)
Abstract
U završnom radu opisan je postupak učenja stabala odlučivanja zanemarujući kompleksne primjere. Tema obuhvaća implementaciju algoritma koji uvažava točnost klasificiranih podataka na isti način kao što to čini i klasifikacijsko stablo. Razumljivost je najvažnija prednost klasifikacijskih stabala u odnosu na većinu drugih klasifikatora1 stoga je vrlo bitna pri izgradnji i implementaciji algoritma. U uvodu se opisuje područje primjene stabala odlučivanja te različitih algoritama koji se koriste kako bi se otkrili što ispravniji, korisniji i razumljiviji modeli iz podataka te je dana sama definicija što je dubinska analiza podataka. Nadalje opisuje se sam problem tj. zašto se gradi novi algoritam koji uvažava dubinu stabla. Predstavljen je kratak opis strojnog učenja (engl. Data Mining), algoritma C4.5 (J48) te je dan primjer korištenja tog algoritma u strojnom učenju. Opis algoritma, kao i sam program koji služi za implementaciju algoritma te programski alat Eclipse također su opisani u daljnjim poglavljima. U prijedlogu algoritma predstavljen je algoritam koji je zapravo bio glavni zadatak ovog završnog rada. Glavnu ideju algoritma predstavlja ocjenjivanje naučenih primjera (engl. Instances) koji se oslanjaju na dubinu lista d u stablu koji je potreban da se primjeri točno klasificiraju u klase kojima pripadaju. Kako bi naučili stabla koristi se već postojeći algoritam C4.5 na svakom slučajnom podskupu atributa.
KeywordsData Mining Classification Tree C4.5 J48 Instances Machine learning Eclipse Comprehensible Trees depth of the tree
Parallel title (English)LEARNING COMPREHENSIBLE TREES DISGARDING COMPLEX INSTANCES
Committee MembersSanda Martinčić-Ipšić (committee chairperson)
Ana Meštrović (committee member)
Lucija Načinović-Prskalo (committee member)
GranterUniversity of Rijeka
Lower level organizational unitsDepartment of Informatics
PlaceRijeka
StateCroatia
Scientific field, discipline, subdisciplineSOCIAL SCIENCES
Information and Communication Sciences
Study programme typeuniversity
Study levelundergraduate
Study programmeInformatics
Academic title abbreviationuniv. bacc. inf.
Genreundergraduate thesis
Language Croatian
Defense date2016-09-08
Parallel abstract (English)
In this final work the proces of learning comprehensible trees by disregarding complex instances is described. The theme includes the implementation of an algorithm which takes into account the accuracy of classified instances the same way as comprehensible classification tree as possible. Comprehensibility is the most important advantage of the classification trees compared to most other classifiers, therefore it is essential for the development and implementation of the algorithm. The introduction describes the domain in which the decision trees and various algorithms are used to discover more correct, useful and understandable models from data and it is given the formal definition of what is data mining. Furthermore the main problem is described and that is, why build a new algorithm that takes into account the depth of the tree. A brief description of Machine learning (Data Mining) is presented, an algortihm C4.5 (J48) and it is given an example the use of that algorithm in Machine learning. Description of the algorithm, the program that is used to implement the algorithm and the software Eclipse are described in further sections. The proposed algorithm is presented in the section “Proposed algortihm” that is actually the main task of this final work. The main algorithm presents grading learned instances which rely on the depth d of the leaf in the tree which is necessary for the instances to be correctly classified in the classes which they belong to. To learn trees, we use the existing algorithm C4.5 on every random subset of attributes. Key words:
Parallel keywords (Croatian)C4.5 J48 klasifikacijsko stablo klasifikator stabla odlučivanja strojno učenje dubina lista primjeri algoritam
Resource typetext
Access conditionNo public access
URN:NBNhttps://urn.nsk.hr/urn:nbn:hr:195:811514
CommitterLea Lazzarich