| dc.description.abstract | Maritime scheduling software is a fairly new and dynamically growing industry. While statistical methods for event impact analysis are well research, they have yet to be applied to this new area. This master thesis investigates the impact of implementing a fleet allocation and scheduling software by a maritime shipping company. Ideas from existing research by Wang et al. (2019) are applied in order to assess event impact on Vessel Weight Utilization, defined as the ratio of cargo weight and the deadweight of the ship, which is a proxy for total cargo carrying capacity. Random Forest is used to predict the "would be" performance of the KPI in a counterfactual scenario in which the software is never introduced. The difference between the actual KPI time series and the "would be" scenario quantifies the software impact.
Furthermore, this paper expands on the existing framework by proposing a way to use Random Forest to make two predictions of the KPI, one for the factual scenario and one for the counterfactual scenario. This allows not only for calculating software impact, but also for prediction distributions to be compared using, among others, kernel density plots and the Kolmogorov–Smirnov test. OLS models are used as a naive benchmark to check the validity of the methods used.
Results suggest that implementing the fleet allocation and scheduling software had a slight effect on the shape of the distribution, but ultimately did not have a visible effect on mean Vessel Weight Utilization over a 2 year period after software implementation. This can be viewed as a positive outcome given that the focus of the Decision Support System during the studied period was on increasing user experience, rather than fleet plan mathematical optimization. The research results indicate that switching to a digital tool marketed as more scalable and increasingly optimization-based does not create a substantial operational risk for the maritime shipping company as measured by the tracked KPI. | en_US |
