This work propeses innovative spatial and spatio-temporal (generalized) linear regression methods for data observed over linear networks. The proposed methods combine a maximum likelihood approach with a regularization penalty involving a differential operator, that has the purpose of controlling the roughness of the estimates. The asymptotic properties of the proposed estimators are studied, proving their consistency and asymptotic normality. The proposed methods are compared to existing methods, such as the geographically weighted regression the rank reduced kriging over linear networks. In these comparative studies the proposed methods appear superior to the alternatives. Moreover, the proposed methods are able to handle data structures that are not covered by existing methods. Finally, we apply the proposed regression estimator to a benchmark dataset containig house price observations of the city of London, UK.
Questo lavoro propone metodi innovativi di regressione lineare spaziale e spazio-temporale (generalizzata) per i dati osservati su reti lineari. I metodi proposti combinano un approccio di massima verosimiglianza con un termine di regolarizzazione che coinvolge un operatore differenziale, che ha lo scopo di controllare la regolarità delle stime. Vengono studiate le proprietà asintotiche degli stimatori proposti, dimostrandone la consistenza e la normalità asintotica. I metodi proposti vengono confrontati con metodi esistenti, come la geographically weighted regression e il rank reduced kriging su reti lineari. In questi studi comparativi i metodi proposti appaiono superiori alle alternative. Inoltre, i metodi proposti sono in grado di gestire strutture di dati che non sono coperte dai metodi esistenti. Infine, applichiamo lo stimatore di regressione proposto a un set di dati di riferimento contenente le osservazioni sui prezzi delle case della città di Londra, Regno Unito.
Spatio temporal models over linear networks
Clemente, Aldo
2021/2022
Abstract
This work propeses innovative spatial and spatio-temporal (generalized) linear regression methods for data observed over linear networks. The proposed methods combine a maximum likelihood approach with a regularization penalty involving a differential operator, that has the purpose of controlling the roughness of the estimates. The asymptotic properties of the proposed estimators are studied, proving their consistency and asymptotic normality. The proposed methods are compared to existing methods, such as the geographically weighted regression the rank reduced kriging over linear networks. In these comparative studies the proposed methods appear superior to the alternatives. Moreover, the proposed methods are able to handle data structures that are not covered by existing methods. Finally, we apply the proposed regression estimator to a benchmark dataset containig house price observations of the city of London, UK.| File | Dimensione | Formato | |
|---|---|---|---|
|
2022_07_clemente_02.pdf
non accessibile
Descrizione: Executive Summary
Dimensione
1.73 MB
Formato
Adobe PDF
|
1.73 MB | Adobe PDF | Visualizza/Apri |
|
2022_07_clemente_01.pdf
non accessibile
Descrizione: Tesi magistrale
Dimensione
3.81 MB
Formato
Adobe PDF
|
3.81 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/191731