This thesis investigates Conformal Prediction as a statistical framework for constructing prediction intervals with finite-sample, distribution-free frequentist coverage guarantees. After presenting its theoretical foundations and main variants, the study highlights how this approach overcomes the limitations of classical predictive methods, particularly in complex and high dimensional settings. Conformal Prediction is then integrated with Machine Learning and time series models. Its performance is empirically assessed through applications in economics and Business Intelligence. The results demonstrate improved uncertainty calibration and enhanced predictive robustness. Overall, Conformal Prediction emerges as a reliable tool for data driven decision support.
Questa tesi analizza la Conformal Prediction come metodologia statistica per la costruzione di intervalli di predizione con garanzie di copertura frequentista finite e distribuzione libere. Dopo averne illustrato i fondamenti teorici e le principali varianti, il lavoro mette in evidenza il superamento dei limiti dei metodi predittivi classici, soprattutto in contesti complessi e ad alta dimensionalità. La Conformal Prediction viene quindi integrata con modelli di Machine Learning e di serie temporali. L’efficacia dell’approccio è valutata empiricamente attraverso applicazioni in ambito economico e di Business Intelligence. I risultati mostrano una migliore calibrazione dell’incertezza e una maggiore robustezza predittiva. La metodologia emerge come strumento affidabile per il supporto alle decisioni data driven.
Conformal Prediction: L’incontro tra Statistica e Machine Learning con applicazioni in Economia e Business Intelligence
MOROSINI, VITTORIA
2025/2026
Abstract
This thesis investigates Conformal Prediction as a statistical framework for constructing prediction intervals with finite-sample, distribution-free frequentist coverage guarantees. After presenting its theoretical foundations and main variants, the study highlights how this approach overcomes the limitations of classical predictive methods, particularly in complex and high dimensional settings. Conformal Prediction is then integrated with Machine Learning and time series models. Its performance is empirically assessed through applications in economics and Business Intelligence. The results demonstrate improved uncertainty calibration and enhanced predictive robustness. Overall, Conformal Prediction emerges as a reliable tool for data driven decision support.| File | Dimensione | Formato | |
|---|---|---|---|
|
Morosini_Vittoria.pdf
accesso aperto
Dimensione
2 MB
Formato
Adobe PDF
|
2 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/20.500.12608/105779