This thesis focuses on the application of Machine Learning algorithms, and in particular Deep Learning, to address and solve problems typical of the industrial field. Specifically, these innovative algorithms, ranging from Random Forest to XGBoost, to Neural Networks, which have shown great developments in recent years, have been tested against four challenging forecasting industrial problems. The first application is about bankruptcy prediction, a task traditionally tackled by statistical methodologies. Furthermore, Deep Learning methodologies are tested against the management of hybrid systems for production planning and control. Also, a statistical based optimization procedure for such system is suggested. Finally, a demand forecasting framework for the fashion industry is proposed. Interesting results have been achieved and indeed, in all occasions, the tested models improved the results obtainable with more classic statistical techniques. Indeed, the Machine Learning techniques have proved to be essential to be able to face problems otherwise difficult to face with classical methodologies.
Questa tesi si concentra sull'applicazione di algoritmi di Machine Learning, ed in particolare di Deep Learning, per affrontare e risolvere problemi tipici del settore industriale. In particolare, tali algoritmi innovativi, tra cui Random Forest, XGBoost e reti neurali, che hanno mostrato grandi sviluppi negli ultimi anni, sono stati testati su quattro impegnativi problemi industriali di previsione. La prima applicazione riguarda la previsione dei fallimenti, un compito tradizionalmente affrontato con metodologie statistiche. Successivamente, le metodologie di Deep Learning sono testate rispetto alla gestione di sistemi ibridi per la pianificazione e il controllo della produzione. Inoltre, viene suggerita una procedura di ottimizzazione basata su tecniche statistiche per tale sistema di gestione. Infine, viene proposto un framework di previsione della domanda per l'industria della moda. Risultati interessanti sono stati raggiunti e, in tutte le occasioni, i modelli testati hanno migliorato i risultati ottenibili con tecniche statistiche più classiche. Le tecniche di Machine Learning si sono infatti dimostrate essenziali per poter affrontare problemi altrimenti difficilmente affrontabili con metodologie classiche.
Industrial applications of machine learning and deep learning algorithms
2020
Abstract
This thesis focuses on the application of Machine Learning algorithms, and in particular Deep Learning, to address and solve problems typical of the industrial field. Specifically, these innovative algorithms, ranging from Random Forest to XGBoost, to Neural Networks, which have shown great developments in recent years, have been tested against four challenging forecasting industrial problems. The first application is about bankruptcy prediction, a task traditionally tackled by statistical methodologies. Furthermore, Deep Learning methodologies are tested against the management of hybrid systems for production planning and control. Also, a statistical based optimization procedure for such system is suggested. Finally, a demand forecasting framework for the fashion industry is proposed. Interesting results have been achieved and indeed, in all occasions, the tested models improved the results obtainable with more classic statistical techniques. Indeed, the Machine Learning techniques have proved to be essential to be able to face problems otherwise difficult to face with classical methodologies.I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14242/134932
URN:NBN:IT:UNIPR-134932