In this work we talk about forecasts and prediction models. We try to learn from the past in order to detect trends and future variations. The main field of research is the commodity stock market and row materials price is the main actor. In particular, we focus on Aluminium, leveraging real industry data. The goal of our analysis is to address the non-stationary component of time series, in order to make more accurate predictions. We develop the research by experimentally comparing different models, from basic models to more complex machine learning models, keeping accuracy and execution time as main metrics. We implement different training architectures and we combine them to get the most efficient predictive model for non-stationary time series. We propose a tree based model performing the best with respect to the others.
In questo lavoro parliamo di previsioni e modelli di predizione. Cerchiamo di imparare dal passato per individuare tendenze lineari e variazioni future. Il principale campo di ricerca è il mercato azionario delle materie prime e il prezzo delle materie prime è l'attore principale. In particolare ci concentriamo sull'alluminio, sfruttando i dati reali del settore. L'obiettivo della nostra analisi è quello di affrontare la componente non stazionaria delle serie temporali, al fine di fare previsioni più accurate. Sviluppiamo la ricerca sperimentale confrontando diversi modelli, dai modelli di base a modelli di apprendimento automatico più complessi, mantenendo accuratezza e tempi di esecuzione come parametri principali. Implementiamo diverse architetture di formazione e valutiamo i risultati per identificare il modello predittivo più efficiente.
A comparative research on non stationary time series prediction : the case of aluminium price
COSTA, LUCA
2018/2019
Abstract
In this work we talk about forecasts and prediction models. We try to learn from the past in order to detect trends and future variations. The main field of research is the commodity stock market and row materials price is the main actor. In particular, we focus on Aluminium, leveraging real industry data. The goal of our analysis is to address the non-stationary component of time series, in order to make more accurate predictions. We develop the research by experimentally comparing different models, from basic models to more complex machine learning models, keeping accuracy and execution time as main metrics. We implement different training architectures and we combine them to get the most efficient predictive model for non-stationary time series. We propose a tree based model performing the best with respect to the others.File | Dimensione | Formato | |
---|---|---|---|
2019_07_Costa.pdf
solo utenti autorizzati dal 12/07/2022
Descrizione: Testo della tesi
Dimensione
2.92 MB
Formato
Adobe PDF
|
2.92 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/10589/149111