Danger Theory or Trained Neural Network – A Comparative Study for Behavioural Detection

Abstract

Danger Theory stipulates a powerful defensive mechanism underpinning the human immune system, which is a sophisticated classification metaphor, namely, Dendritic Cell Algorithm (DCA) which has been demonstrated in many real-life applications. In this paper, the DC-inspired metaphor is adopted in the domain of the behavioural detection for field operation training. This signal-based classification algorithm empowers with a robust learning capability and self-organizing control mechanism, whereby the 'danger signals' correspond to the safety and procedural related activities, are being differentiated from the information which are captured in immersive virtual environments including the Training of Ramp Operations in Virtual Environment (TROVE) and Detective Boulevard supported by the imseCAVE virtual reality system. As such, the performance of the trainees can be assessed and enumerated by the DCA autonomously, that can improve the quality of assessment made by the trainers or coaches. In a benchmark study, the operational training of aircraft door handling is considered, in particular to study the behaviours in operational procedures and safety concerns. According to the experimental results, Artificial Neural Network (ANN) outperformed DCA in the given domain with respect to the performance of classification accuracy and run time.