Insider threats pose an unavoidable risk in information security. These threats, malicious or not, leave industries and organizations susceptible to significant losses, ranging from customer information to corporate financial records. As this is a continuous threat whose victim count is ever-growing, it provides an opportunity to explore solutions for identifying these malicious events before its effects materialize. Given the success of machine learning techniques in providing powerful solutions to previously unsolvable problems, our research team is focusing on the application of neural networks on a large distributed transaction dataset made available by the Los Alamos National Laboratory. We have applied recurrent neural network algorithms on this dataset which consists of over a billion transactions to predict the next actions of different users which will lead us to detection of potential insider threats. The visualization project aims at using extracted patterns, developed by the aforementioned models, to further analyze the information by means of visualization techniques provided by the Neo4j graph database and its graphical toolset.

Such visual techniques and exploration activities will be complementary to the analytical aspects of machine learning process as they provide deep insight into the interconnections between the sequence of data in the event log. We will present different graphs with filtering capabilities to declutter the extracted patterns from machine learning stage which allows for an interactive exploration experience for the security administration of complex and large systems.