Context-aware Purchasing Assistant

In economy, it is important to use the storage as efficient as possible.  Storage space is "unprofitable" for the business because it is not a productive space. For articles which need special storage, the storage costs the business even more. This results in storage optimization. If the storage is too small, there is not enough space for some articles and they cannot be sold even if the customers demand it. If the business stores too much of an article, it wastes space and some articles can perish. These losses can be minimized with an accurate sale forecast. While old forecast techniques just use the history, new methods can take external factors, as the price of an article, into account. Because of the human's missing trust in new systems, these forecast techniques have to be made interpretable for humans. This thesis analyzes a new neural network forecasting technique with Gated Recurrent Units (GRU), compares it to older forecasting techniques and introduces a new interpretation technique in order to make the GRU-based time series prediction interpretable for users.