Accurate ship valuation can encourage transparency and reliability in the shipping industry. In this age driven by artificial intelligence; however, deep learning approaches have not yet taken root in ship valuation. Despite the significant achievements of deep learning algorithms in the field of unstructured data such as computer vision, the same cannot be said for the structured data-dominant areas, including the shipping industry. Neural networks (NNs), the most common algorithms for implementing deep learning, are known not to have a relative advantage in handling structured data, particularly in processing categorical data. The inefficiency of NNs for processing categorical data significantly degrades their performance when categorical data occupy a significant portion of a dataset. In this study, we employed a NN to estimate second-hand ship prices. Its architecture was specified using entity embedding layers to enhance the performance of the network when categorical variables were highly cardinal. Experimental results demonstrated that the information contained in categorical data can be efficiently extracted and fed into a NN using the entity embedding technique, thereby improving the prediction accuracy for ship valuation. The network architecture specified in this study can be applied in wider valuation areas where categorical data are prevalent.