Estimating transport lead time parameters in Scania supply chain

Sammanfattning

As the usage of machine learning techniques increases in modern manufacturing industries, this also presents new ways to optimize supply chain processes. This thesis investigates the effectiveness of three different machine learning methods for estimating transport lead time parameters. The study is conducted in collaboration with Scania, where accurate transport lead time estimates are important for efficient supply chain planning and reliable deliveries to customers. In many cases, current transport lead time parameters are static and not updated regularly, creating substantial differences to the naturally versatile real-life operations. This may lead to inefficiencies such as delays and wasted resources. The data used in this study comes from Scania's internal systems and includes information about order deliveries, such as dates, delivery country and production status. Data preprocessing was done to clean the dataset: handling missing or illogical values and removing outliers. Before applying the models, a multicollinearity analysis was performed to understand the relationships between variables. The machine learning methods applied in this study are linear regression, k-nearest neighbors, and random forest. These models are used to predict transport lead time based on selected features Delivery country, Weekday, Month, Production status, ZIP codes and Distance. The models are evaluated using error calculation, cross-validation, and mean predictions across groups of ZIP codes, and based on this assessment random forest is shown to be the most fitting of the chosen models.