Geograafiliste Andmete Mudelipõhine Rühmitamine
Files
Date
2012
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Tartu Ülikool
Abstract
Töö eesmärk oli disainida ja realiseerida algoritme teatud tüübiga andmete analüüsimiseks. Tegu on geograafiliste koordinaatidega annoteeritud sündmustega (pikkus ja laius). Algoritmide põhiülesanne on ennustada geograafilisi piirkondi, kus sama tüübiga sündmused toimuvad suure tõenäosusega ka tulevikus. Me vaatlesime kahte tüüpi andmeid. Esimene tüüp on kahemõõtmelise normaaljaotusega sündmused, mis paiknevad ümber keskset punktallikat. Teine tüüp on sündmused, mis on jaotatud mööda keskset joonelõiku. Näiteks liiklusõnnetuste toimumiskohad paiknevad mööda maanteesid, mida saab modelleerida sirgete lõikude kogumina. Esimene andmetüüp on laialt esitatud kirjanduses, ning antud töös kirjeldatakse hästi tuntud sellekohased algoritmid. Teise andmetüübi analüüsimiseks sobivate algoritmide arendus ja esitamine oli aga selle töö peamine panus.
This work was motivated by the Virtual Situation Room project. Its purpose is to predict the occurrence of some destructive events based on existing data. The goal of this thesis was to design and implement algorithms useful for analyzing geo-tagged events. We focus on two different kinds of data. The first type is data distributed normally around central point sources which is a very common model in literature and in practice. The second type and the main focus of this work is data distributed along line segment sources. For example traffic accidents in rural area are likely to be distributed along highways which can be modeled as sequences of segments. In each case the goal is to identify regions where the events of the same type are likely to occur in the future. We accomplish this by applying model-based clustering techniques to our existing data observations.
This work was motivated by the Virtual Situation Room project. Its purpose is to predict the occurrence of some destructive events based on existing data. The goal of this thesis was to design and implement algorithms useful for analyzing geo-tagged events. We focus on two different kinds of data. The first type is data distributed normally around central point sources which is a very common model in literature and in practice. The second type and the main focus of this work is data distributed along line segment sources. For example traffic accidents in rural area are likely to be distributed along highways which can be modeled as sequences of segments. In each case the goal is to identify regions where the events of the same type are likely to occur in the future. We accomplish this by applying model-based clustering techniques to our existing data observations.