Data analytics for optimal container shipping

Ζήσης, Κωνσταντίνος; Zisis, Konstantinos

dc.contributor.author	Ζήσης, Κωνσταντίνος	el
dc.contributor.author	Zisis, Konstantinos	en
dc.date.accessioned	2025-01-14T11:53:38Z
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/60729
dc.identifier.uri	http://dx.doi.org/10.26240/heal.ntua.28425
dc.rights	Default License
dc.subject	Logistics	en
dc.subject	AI	en
dc.title	Data analytics for optimal container shipping	en
heal.type	bachelorThesis
heal.classification	Supply Chain	en
heal.dateAvailable	2026-01-13T22:00:00Z
heal.language	en
heal.access	embargo
heal.recordProvider	ntua	el
heal.publicationDate	2024-07-17
heal.abstract	This thesis investigated the feasibility of accurately predicting delays in container shipping for big multinational companies. Two projects were completed, different in scope but related by their common field. Both these projects were completed with the help and the data of Sappi, which is a big multinational company focusing on forestry products and papers. The first project was a user experience (UX) project on a supply chain control tower in order to find out the needs and requirements of supply chain employees and managers. The main result of this project was that Sappi’s supply chain needed more predictive analytics capability; hence in the second project a predictive model was developed to predict whether a container shipment would be late. The research employed a four-stage methodology: describe, predict, diagnose, and prescribe. In the describing stage, relevant factors influencing container shipping delays were identified. The predicting stage focused on developing machine learning models to predict these delays. Target encoding and label encoding were compared for representing categorical features, and the performance of various algorithms (XGBoost, Random Forest, Neural Networks, Logistic Regression, as well as a knowledge-based model) was evaluated. The diagnosing stage aimed to understand the root causes of delays based on model predictions. Finally, the prescribing stage is done by incorporating some rules as to how these tools should be used. Creating a fully automated supply chain control tower is beyond the scope of this thesis. The key findings of the research were that machine learning models can effectively predict delays in container shipping for big multinational companies with the random forest model achieving the highest accuracy in predicting delays. Moreover, as expected, target encoding showed a slight advantage over label encoding for representing categorical features in this specific case. Unfortunately, the analysis revealed that most of the features that are most useful in predicting delays are not directly affected by Sappi. However, Sappi can use these findings as negotiating tools, both with customers (e.g. timing their orders more efficiently) and with their partners (e.g. negotiating prices based on performance). To conclude, this thesis’ key outputs are the predictive model that can predict delayed containers and insights as to why containers are getting delayed. This research contributes to the field of logistics by demonstrating the potential of machine learning and predictive algorithms for proactive delay management in container shipping. The findings provide valuable insights for companies that rely on port-to-port shipping seeking to optimize their supply chains and mitigate the negative impacts of delays. The methodology employed in this thesis also holds promise for adaptation in other domains, laying the groundwork for future research exploring its transferability to diverse areas.	en
heal.advisorName	Ρεντιζέλας, Αθανάσιος	el
heal.committeeMemberName	Τόλης, Αθανάσιος	el
heal.committeeMemberName	Πόνης, Σταύρος	el
heal.committeeMemberName	Ρεντιζέλας, Αθανάσιος	el
heal.academicPublisher	Εθνικό Μετσόβιο Πολυτεχνείο. Σχολή Μηχανολόγων Μηχανικών. Τομέας Βιομηχανικής Διοίκησης και Επιχειρησιακής Έρευνας	el
heal.academicPublisherID	ntua
heal.numberOfPages	79 σ.	el
heal.fullTextAvailability	false
heal.fullTextAvailability	false