The analysis and prediction of the risks and causes of railway freight accidents can help to formulate accident prevention measures, thereby ensuring the safety of freight railway operations. In order to achieve the prediction of the evolution results of railway freight accidents and reduce the probability of accidents from the source, this paper proposes a prediction method for the evolution results of railway freight accidents based on the K2 scoring algorithm and the graph convolutional neural network (GCN) model. First, based on the data from the Federal Railroad Administration (FRA) railroad equipment accident database, this paper classifies the causes of accidents into five major categories from the perspectives of human, machine, and environment for a total of five categories of accidents, including derailment, collision, level crossing, and foreign object intrusion. Second, based on the K2 scoring algorithm, the network graph was constructed from the overall perspective considering various influencing factors such as operating environment and using structured scenario information. Then the railroad freight accident outcome prediction model was constructed based on the graph convolutional neural network model. Finally, 8910 data are used for classification prediction analysis. The experimental results show that the proposed method has a good prediction effect and is generally applicable, which can provide methods and tools to support accident investigation and cause analysis.