In hospital emergency rooms, where workloads are inherently inhomogeneous, predicting those workloads as accurately as possible plays an essential role in employee shift planning. These workloads are not random but the result of the complicated interaction of environmental influences on human health. Artificial neural networks can use data to identify correlations between those environmental influences and hospital workloads, predicting future workloads based on new data. While the total workload of a hospital emergency room is interesting and already researched, no work tries to predict multiple different categories of diagnosis to identify the staff that has to be ready for any given time. This thesis aims to show that artificial neural networks can predict the workloads of different diagnosis categories. We use historical data on multiple environmental influences and the emergency room of the Universitätsklinikum in Freiburg, Germany, to train four different artificial neural networks. While the metrics show a promising result, the networks have problems accurately predicting outliers, like extremely high and low workloads. This makes the networks a reasonable basis for further research but, in their current state, irrelevant for a real-life application.