Insider Threat Detection Using Neural Networks: A Statistical Comparison of the CERT Insider Threat Database
Open Access DepositedThis research compares the efficacy of various classifier TensorFlow Neural Networks (NNs) at detecting Insider Threats to guide designers of security solutions on the selection of appropriate NN algorithms for Insider Threat detection. The praxis utilized a modified version of the CERT r4.2 dataset provided by David A. Noever. Fifteen TensorFlow NN models were compared, as were different rescaling techniques and epoch counts.A Kruskal-Wallis test and Dunn’s post hoc test were used to compare NN models, which revealed differences in performance. The top-performing model on average was non-rescaled ELM-klm 10, which achieved an accuracy of 0.9804. That is in contrast with the top performer of Noever’s research, Random Forest, which achieved an accuracy of 0.9835. It should also be noted that this praxis utilized 10 risk indicators to achieve this performance by saving processing time with only a small loss of accuracy. Furthermore, there was a significant difference across rescaling techniques such as Standardize, Normalized, Non-Rescaled or Robust, as was also the case for the utilization of 10 Epoch counts or 35 Epoch counts. While these findings align with previous literature in terms of showcasing different levels of accuracy across models using various techniques, the specific outcomes differ when compared to other literature, with accuracy differences being small. Future research could expand on this and similar studies by testing other datasets, as well as evaluating additional NN models and platforms.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.