FDA is a continuously developing field, but even though its influence is constantly growing, there is still an open concept: the uncertainty quantification in forecasting. As a matter of fact, given a valid prediction set of level 1−α, a new curve with a different behaviour from the main dataset’s attitude may be incorrectly predicted. This phenomenon is related to a different covariance structure. However from an application point of view this is truly a problem to be faced. Starting from the Split Conformal algorithm in a multivariate functional setting, a possible solution consists in replacing functional components with differential quantities of the original dataset. As a result, in this thesis the main Split Conformal properties are firstly examined as the covariance structure between components varies; then the simultaneous prediction of the dataset and its differential quantities, such as derivatives, are deeply discussed. The method is investigated and analyzed through Monte Carlo simulations and a realworld application in the field of urban mobility.
FDA è un campo in continuo sviluppo. Nonostante stia prendendo sempre più piede, un problema ancora aperto riguarda il concetto di previsione. Infatti dato un intervallo di previsione di livello 1 − α, un nuovo dato con una distribuzione diversa dal dataset può essere erroneamente considerato come prevedibile. Il motivo principale di questa previsione sbagliata è una questione di diversa struttura di covarianza. Tuttavia dal punto di vista applicativo è un problema che va affrontato. Partendo dal metodo Split Conformal applicato a dati funzionali multivariati, una possibile soluzione consiste nel considerare simultaneamente grandezze differenziali del dataset. Di consequenza in questa tesi esaminiamo le principali proprietà del metodo Split Conformal al variare della struttura di covarianza tra le componenti e successivamente discuteremo della previsione simultanea di un nuovo dato considerando il dataset stesso e le sue derivate. Questo approccio viene analizzato attraverso diverse simulazioni Monte Carlo ed applicato ad un caso studio legato alla mobilità del bike-sharing.
Multi-aspect conformal prediction bands for functional data
Rota, Francesca
2021/2022
Abstract
FDA is a continuously developing field, but even though its influence is constantly growing, there is still an open concept: the uncertainty quantification in forecasting. As a matter of fact, given a valid prediction set of level 1−α, a new curve with a different behaviour from the main dataset’s attitude may be incorrectly predicted. This phenomenon is related to a different covariance structure. However from an application point of view this is truly a problem to be faced. Starting from the Split Conformal algorithm in a multivariate functional setting, a possible solution consists in replacing functional components with differential quantities of the original dataset. As a result, in this thesis the main Split Conformal properties are firstly examined as the covariance structure between components varies; then the simultaneous prediction of the dataset and its differential quantities, such as derivatives, are deeply discussed. The method is investigated and analyzed through Monte Carlo simulations and a realworld application in the field of urban mobility.| File | Dimensione | Formato | |
|---|---|---|---|
|
2022_07_Rota.pdf
non accessibile
Descrizione: Testo Tesi
Dimensione
2.48 MB
Formato
Adobe PDF
|
2.48 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/10589/190429