In an e-commerce company, the packaging process is an essential part of the delivery process. If the appropriate packaging size is proposed to the warehouse worker during this process, time can be saved. In addition, the prediction enables the estimation of the delivery volume in advance. This master's thesis was carried out in cooperation with the company niceshops GmbH, which develops online shops in various product segments. The goal of this thesis is to examine whether it is possible to predict the packaging size of a delivery using a machine learning model based on the data of historical deliveries. The relevant data for the prediction was analyzed and prediction models were trained with different algorithms such as k-nearest neighbor, random forest or backpropagation. The models were evaluated for each online shop. The accuracy of the predictions made by these models varies from shop to shop. This is due to the different variety of products in the shops as well as the data quality regarding the selection of the correct packaging size in earlier deliveries. Overall, the random forest algorithm provided the best results. A corresponding model was integrated into the packaging process of one of the shops and is used to propose one out of over 40 different packaging sizes to the warehouse worker.