Change detection and adaptation in multi-target prediction on data streams

Stevanoski, Bozhidar (author), Džeroski, Sašo (mentor)

An essential characteristic of data streams is the possibility of occurrence of concept drift, i.e., change in the distribution of the data in the stream over time. The capability to detect and adapt to changes in data stream mining methods is thus a necessity. While methods for multi-target prediction on data streams have recently appeared, they have largely remained without such capability. In this thesis, we develop methods for change detection and adaptation in the context of incremental online learning of decision trees for multi-target regression. One of the approaches we propose is ensemble based, while the other uses the Page-Hinckley test. We extend both to also operate in the context of semi-supervised learning from partially labeled data. We perform an extensive evaluation of the proposed methods on real-world and artificial data streams and show their effectiveness, also on a case study from spacecraft operations.

• Language: English
• Keywords: machine learning, data streams, change detection and adaptation, multi-target prediction, multi-target regression
• Work type: Master's thesis/paper
• Typology: 2.09 - Master's Thesis
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2021
• PID: 20.500.12556/RUL-131001
• COBISS.SI-ID: 78733059
• Publication date in RUL: 21.09.2021

Secondary language

• Language: Slovenian
• Title: Zaznavanje in prilagajanje spremembam pri večciljnem napovedovanju na podatkovnih tokovih

Abstract

Bistvena značilnost podatkovnih tokov je možnost pojava konceptnega premika, to je spremembe v porazdelitvi podatkov v toku skozi čas. Sposobnost odkrivanja in prilagajanja spremembam pri metodah podatkovnih tokov je zato nujna. Čeprav so se pred kratkim pojavile metode za večciljno napovedovanje na podatkovnih tokovih, le te so še vedno brez takšnih zmogljivosti. V tej nalogi razvijemo metode za odkrivanje sprememb in prilagajanje le-tem v kontekstu inkrementalnih dreves za večciljno regresijo. Eden od pristopov, ki jih predlagamo, temelji na ansambliskih metodah, drugi pa uporablja Page-Hinckleyjev test. Oba pristopa razširjimo tudi za delo v okviru polnadzorovanega učenja iz delno označenih podatkov. Opravimo obsežno evalvacijo predlaganih metod ja resničnih in umetnih tokovih podatkov ter pokažemo učinkovitost teh metod tudi na študiji primera s podroćja upravjanja vesoljskih plovil.

• Keywords: strojno učenje, podatkovni tokovi, zaznavanje sprememb in prilagajanje, večciljno napovedovanje, večciljna regresija