Postupci strojnog učenja za popravljanje točnosti klasifikacije manjinskih klasa kod nebalansiranih skupova podataka

undergraduate thesis

• Title (english): Machine Learning Methods for Improving Classification Accuracy of Minority Classes on Unbalanced Datasets
• Author: Marko Josipović
• Mentor: Alan Jović (mentor)
• Committee member: Alan Jović (predsjednik povjerenstva)
• Committee member: Nikolina Frid (član povjerenstva)
• Committee member: Igor Stančin (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2019-07-12, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Ovaj rad uspoređuje utjecaj postupaka za popravljanje točnosti i njihovih kombinacija na klasifikaciju manjinskih klasa kod nebalansiranih skupova podataka. Za potrebe eksperimentalnog vrednovanja razvijena je web aplikacija. Na temelju rezultata eksperimentalnog vrednovanja donesen je zaključak da odabir postupaka za popravljanje točnosti ovisi o karakteristikama skupa nad kojim se obavlja klasifikacija. Postupci ponovnog uzorkovanja koji su postigli najbolje rezultate nad skupovima podataka odabranima za vrednovanje su: SMOTE te kombinacije SMOTE+ADASYN, SMOTE+Random Under, SMOTE+ADASYN+NearMiss, SMOTE+ADASYN+Random Under, SMOTE+NearMiss+Random Under i kombinacija sva 4 postupka. Osim postupaka ponovnog uzorkovanja uspoređena su dva algoritma klasifikacije, rezultati pokazuju da je za klasifikaciju nebalansiranih skupova podataka ansambl klasifikatora balansirane slučajne šume prikladniji od algoritma knajbližih susjeda.

Abstract (english)

This paper compares the influence of methods and combinations of methods for improving classification accuracy of minority classes on unbalanced datasets. Web application has been developed for the purpose of experimental evaluation. Based on the results of the experimental evaluation, it was concluded that the choice of methods depends heavily on the characteristics of datasets on which the classification was performed. Resamplingmethodsthatachievedthebestresultsare: SMOTEandcombinations SMOTE+ADASYN, SMOTE+Random Under, SMOTE+ADASYN+NearMiss, SMOTE+ADASYN+RandomUnder,SMOTE+NearMiss+RandomUnderandthecombination of all 4 methods. In addition to resampling methods, the paper compares two classification algorithms. The results show that Balanced Random Forest ensemble is better suited for classification of unbalanced datasets than k Nearest Neighbors classifier.

• Keywords: poduzorkovanje; naduzorkovanje; ansambl klasifikatora; matrica zabune; mjera F1
• Keywords (english): undersampling; oversampling; ensemble of classifiers; confusion matrix; F1 score
• Language: croatian
• URN:NBN: urn:nbn:hr:168:156417
• Study programme: Title: Computing; Study programme type: university; Study level: undergraduate; Academic / professional title: sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva (sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2020-01-07 18:44:28