Document Type

Conference Proceeding




Faculty of Computing, Health and Science


School of Computer and Security Science / Artificial Intelligence and Optimisation Research Centre




This is an Author's Accepted Manuscript of: Fanchao, Z., Decraene, J., Low, M., Cai, W., Zhou, S., & Hingston, P. F. (2011). High-dimensional objective-based data farming. Paper presented at the IEEE Symposium on Computational Intelligence for Security and Defense Applications. IEEE. Paris, France. Available here

© 2011 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.


In objective-based data farming, decision variables of the Red Team are evolved using evolutionary algorithms such that a series of rigorous Red Team strategies can be generated to assess the Blue Team's operational tactics. Typically, less than 10 decision variables (out of 1000+) are selected by subject matter experts (SMEs) based on their past experience and intuition. While this approach can significantly improve the computing efficiency of the data farming process, it limits the chance of discovering “surprises” and moreover, data farming may be used only to verify SMEs' assumptions. A straightforward solution is simply to evolve all Red Team parameters without any SME involvement. This modification significantly increases the search space and therefore we refer to it as high-dimensional objective-based data farming (HD-OBDF). The potential benefits of HD-OBDF include: possible better performance and information about more important decision variables. In this paper, several state-of-the-art multi-objective evolutionary algorithms are applied in HD-OBDF to assess their suitability in terms of convergence speed and Pareto efficiency. Following that, we propose two approaches to identify dominant/key evolvable parameters in HD-OBDF - decision variable coverage and diversity spread.



Access Rights




Link to publisher version (DOI)