Nadogradnja raspodijeljenih modela geografski ponderirane i grupirane regresije

master's thesis

• Title (english): Extending Distributed Models of Geographically Weighted and Clustered Regression
• Author: David Buntak
• Mentor: Krešimir Pripužić (mentor)
• Committee member: Krešimir Pripužić (predsjednik povjerenstva)
• Committee member: Siniša Sovilj (član povjerenstva)
• Committee member: Jurica Babić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2021-07-09, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

U ovom diplomskom radu opisana je nadogradnja modela geografski grupirane regresije dodatnim algoritmima grupiranja DBSCAN i mean shift te modela geografski ponderirane regresije s tri jezgrene funkcije: Laplaceova, Epanechnikova i kosinusna. Nadogradnje su izvedene u programskom okviru Apache Spark. Prikazane su evaluacije u obliku vremena izvođenja i/ili pogreški predviđanja koristeći implementirane nadogradnje nad skupom podataka koji se sastoji od 90275 opažanja. Također provedena je usporedba algoritama grupiranja DBSCAN, mean shift i KMeans. Uz općenitu usporedbu načina rada navedenih algoritama uspoređeni su i rezultati grupiranja na već spomenutom skupu podataka te pogreške predviđanja geografski grupirane regresije. Pokazalo se da DBSCAN ima manju pogrešku u tri od četiri korištene metrike od preostala dva algoritma, dok mean shift pokazuje nešto bolje rezultate od algoritma KMeans. Također, provedena je i evaluacija implementiranih jezgrenih funkcija. Evaluacija je pokazala da nema velike razlike u točnosti predviđanja geografski ponderirane regresije za različite jezgrene funkcije.

Abstract (english)

This thesis describes the upgrade of the geographically clustered regression model with an additional DBSCAN grouping algorithm and the geographically weighted regression model with three kernel functions: Laplacian, Epanechnikov and cosine. The upgrades were performed in the Apache Spark software framework. Evaluations in the form of runtime and / or prediction errors are presented using implemented upgrades over a data set consisting of 90275 observations. A comparison of the DBSCAN, mean shift and KMeans grouping algorithms was also performed. In addition to a general comparison of the modes of operation of these algorithms, the results of grouping on the already mentioned dataset and the prediction errors of geographically clustered regression were also compared. DBSCAN was shown to have a smaller error in three of the four metrics used than the remaining two algorithms, while mean shift shows slightly better results than the KMeans algorithm. Also, an evaluation of the implemented kernel functions was performed. The evaluation showed that there is no large difference in the accuracy of predicting geographically weighted regression for different kernel functions.

• Keywords: geografski ponderirana regresija; geografski grupirana regresija; geoprostorni podaci; DBSCAN; KMeans; mean shift; jezgrena funkcija; Apache Spark; algoritmi grupiranja
• Keywords (english): geographically weighted regression; geographically clustered regression; geospatial data; DBSCAN; KMeans; mean shift; kernel function; Apache Spark; clustering algorithms
• Language: croatian
• URN:NBN: urn:nbn:hr:168:439911
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-01-13 14:39:57