Application of machine learning clustering techniques for performance analysis in ocean freight transport: a Ferrero case study

This thesis applies machine learning clustering techniques to the analysis of ocean freight transport flows, with the aim of identifying patterns and inefficiencies. The method is applied to the international logistics network of the Ferrero Group. The primary objective of this research is to identify inefficiencies within a portion of the supply chain, i.e. the distribution process, by examining three crucial performance indicators: the total storage days products spend in warehouses, the overall lead times from production to shipment, and the associated risk of aging stock that might lead to expiration. To this aim, two clustering methods, K-Means and DBSCAN, were applied, in order to group products and highlight similar logistics behavior. The missing values were handled via median imputation, numerical variables were standardized and categorical attributes were label-encoded. Then K-Means and DBSCAN were executed: K-Means chose k = 2 (silhouette 0.504), yielding a dominant baseline cluster (~95%) and a compact hotspot (~5%) with markedly elevated storage days (~660), longer lead times (~32) and high ageing exposure (~1,123). DBSCAN was a density-based check and validated the hotspot, revealing dense local pockets and outliers associated with lanes, temperature classes or pack types. In a simple case, reducing the average storage in the hotspot from about 660 days to 330 takes the network-wide average back by 12 percent, hence where a particular tactic (e.g., beginning a new booking earlier) is more likely to pay the greatest payoff. The work demonstrates that this critical point is composed of the smallest percentage of SKUs products whose excessive time in the warehouse and high exposure to shelf-life expiration risk disproportionately drive the overall system's inefficiency. This led to the key conclusion that, for this specific small segment of products, focusing on targeted improvement strategies, such as revising inventory policies or optimizing shipping schedules, can yield significant gains for the company's logistics and supply chain performance.