One of the tasks of stock management is determine the safety stock of a product. The aim of the safety stock is to avoid out-of-stock situations when demand increases unexpectedly. In most of manufacturing companies, the safety stock of a material is established by supply planners and they usually do not have a clear method to do it, they simply decide it from previous experience. Thus, the objective of this project is to perform a neural network capable of substituting this task developed by a human brain. To do so, an international manufacturing company provided real data to test the results. The research is divided into two parts. In the first trial a linear regression model is fitted to find the significant variables of the data given, and then a neural network is implemented with only the relevant inputs. Secondly, several supply planners are interviewed in order to adopt the variables they use to decide the safety stock as inputs of the neural network, in addition, the dataset is separated into three groups of products according to their similarity, and one neural network is implemented for each group of products. The results obtained in both parts are not good enough, that is, the neural networks built cannot replace the job done by a supply planner. However, it is found that the more similar the products of the dataset are, the easier it is for the neural network to predict their safety stock. In fact, the best neural network performed can accurately determine the safety stock of some materials, even though the total error is too high to consider capable of substituting the decision making of a human brain.
