delo diplomskega seminarja
Sara Bizjak (Author), Ljupčo Todorovski (Mentor)

Abstract

Ideja strojnega učenja ansamblov je zgraditi napovedni model z združevanjem večih modelov, kar pripomore k manjšanju napovedne napake. Ena ključnih komponent ansambla je funkcija za kombiniranje napovedi osnovnih modelov. V diplomskem delu obravnavamo dva tipa funkcij za kombiniranje napovedi klasifikacijskih modelov. Prvi je večinsko glasovanje, kjer vsi osnovni modeli enako prispevajo k napovedi ansambla. Drugi pa je uteževanje prispevka osnovnih modelov na osnovi njihove zmogljivosti. Ti dve funkciji kombiniranja implementiramo v programskem jeziku R in ju primerjamo na izbrani podatkovni množici.

Keywords

strojno učenje;nadzorovano strojno učenje;klasifikacija;homogeni ansambli;naključni gozd;kombiniranje napovedi;uteževanje na osnovi zmogljivosti;

Data

Language: Slovenian
Year of publishing:
Typology: 2.11 - Undergraduate Thesis
Organization: UL FU - Faculty of Administration
Publisher: [S. Bizjak]
UDC: 004.8
COBISS: 58738435 Link will open in a new window
Views: 1351
Downloads: 346
Average score: 0 (0 votes)
Metadata: JSON JSON-RDF JSON-LD TURTLE N-TRIPLES XML RDFA MICRODATA DC-XML DC-RDF RDF

Other data

Secondary language: English
Secondary title: Techniques for combining predictions in machine learning ensembles
Secondary abstract: Machine learning of ensembles aims at reducing the predictive error by integrating multiple models into a single one. One of the key components of algorithms for ensemble learning is combining predictions of the base models. In the thesis, we take a closer look at two functions for combining predictions. The first is majority voting, where all the base models contribute equally to the ensemble prediction. The other is performance weighting, where the contribution of a base model to the ensemble prediction is proportional to the model performance. Combination functions are also implemented in R and tested on a selected data set.
Secondary keywords: machine learning;supervised machine learning;classification;homogeneous ensembles;random forest;combining predictions;performance weighting;
Type (COBISS): Final seminar paper
Study programme: 0
Embargo end date (OpenAIRE): 1970-01-01
Thesis comment: Univ. v Ljubljani, Fak. za matematiko in fiziko, Oddelek za matematiko, Matematika - 1. stopnja
Pages: 31 str.
ID: 12044725
Recommended works: