A predictive User behaviour analytic Model for Insider Threats in Cyberspace

Abstract

Insider threat in cyberspace is a recurring problem since the user activities in a cyber network are often unpredictable. Most existing solutions are not flexible and adaptable to detect sudden change in user’s behaviour in streaming data, which led to a high false alarm rates and low detection rates. In this study, a model that is capable of adapting to the changing pattern in structured cyberspace data streams in order to detect malicious insider activities in cyberspace was proposed. The Computer Emergency Response Team (CERT) dataset was used as the data source in this study. Extracted features from the dataset were normalized using Min-Max normalization. Standard scaler techniques and mutual information gain technique were used to determine the best features for classification. A hybrid detection model was formulated using the synergism of Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU) models. Model simulation was performed using python programming language. Performance evaluation was carried out by assessing and comparing the performance of the proposed model with a selected existing model using accuracy, precision and sensitivity as performance metrics. The result of the simulation showed that the developed model has an increase of 1.48% of detection accuracy, 4.21% of precision and 1.25% sensitivity over the existing model. This indicated that the developed hybrid approach was able to learn from sequences of user actions in a time and frequency domain and improves the detection rate of insider threats in cyberspace.