In recent times, the ever-increasing amount of available data has made dimensionality reduction one of the most important statistical tools. Principal component analysis is the most widespread and well-established technique, thanks to its simplicity and clear interpretability. However, in its classic formulation, PCA has some significant limitations, such as the difficulty in choosing the number of components and its weaknesses in handling missing data. To overcome these limitations, alternative approaches have been proposed, including probabilistic PCA and Bayesian PCA. The aim of this thesis is to compare the three approaches, with a particular focus on the Bayesian one, both theoretically and practically, highlighting the conditions under which each method is more suitable for real-world data with very different characteristics.
In tempi recenti la mole sempre maggiore di dati a disposizione ha reso la riduzione della dimensionalità uno tra gli strumenti statistici più importanti e l’ analisi delle componenti princiapli è la tecnica più diffusa e consolidata, grazie alla sua semplicità e alla chiara interpretabilità. Tuttavia, nella sua formulazione classica, la PCA presenta alcune limitazioni rilevanti, come la difficolta nella scelta del numero di componenti e la sua debolezza nella gestione dei dati mancanti. Per superare questi limiti, sono stati proposti approcci alternativi tra cui la PCA probabilistica e la PCA bayesiana. L’obiettivo di questa tesi è confrontare i tre approcci, con particolare attenzione a quello bayesiano, sia sul piano teorico che su quello applicativo, mettendo in risalto le condizioni in cui ciascun metodo risulta più adeguato rispetto a dati reali con caratteristiche molto diverse.
PCA Bayesiana: Un Confronto Teorico e Applicativo con i Metodi Classici
BIZZOTTO, MILO
2025/2026
Abstract
In recent times, the ever-increasing amount of available data has made dimensionality reduction one of the most important statistical tools. Principal component analysis is the most widespread and well-established technique, thanks to its simplicity and clear interpretability. However, in its classic formulation, PCA has some significant limitations, such as the difficulty in choosing the number of components and its weaknesses in handling missing data. To overcome these limitations, alternative approaches have been proposed, including probabilistic PCA and Bayesian PCA. The aim of this thesis is to compare the three approaches, with a particular focus on the Bayesian one, both theoretically and practically, highlighting the conditions under which each method is more suitable for real-world data with very different characteristics.| File | Dimensione | Formato | |
|---|---|---|---|
|
Bizzotto_Milo.pdf
accesso aperto
Dimensione
3.35 MB
Formato
Adobe PDF
|
3.35 MB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/115383