Anomaly detection with negative selection and graphs of user behavior profiles

Abstract

This work proposes the application of a Negative Selection Algorithm inspired in Artificial Immune Systems and the use of Graphs. This approach allows to identify anomalous activities based on user's behavior, during the execution of tasks, as well as at the end of the user session. The main contributions of this work are: creation of a graph-based user profile for the detection and prediction of anomalies in execution tasks. In addition, the generation of anomalous behaviors for the detection of irregularities at the end of the user session. The article presents in detail the development and the acceptable result in the detection of fraudulent tasks, showing an optimum precision and a low percentage of false positives. © AISTI 2016.