Simulation for a human inspired trust network. Master thesis 2017
##Abstract
Car2Car communication has the potential to increase traffic safety by expanding the sensor view from only internal sensors to external sources. One of the key success factors for such Car2Car communication is trustworthy and reliable information. Current centralized trust concepts are limited as they expose the privacy of the driver. Decentralized, human inspired trust models avoid a central authority and therefore increase privacy.
Based on a theoretical overview of different trust concepts and components, I develop a probabilistic simulation, which builds a trust network based on the feedback of the interaction between agents. I test the influence of trust knowledge, recommendation and trust facets dimensions in different scenarios.
The results of the simulated scenarios show that human inspired trust can increase the rate of correct decisions by 32%. In low density networks the increase of recommendation depth improves the success rate by 37%. If agents can send a certainty along with their statement, the success improves with up to 30%.
The simulation results show that a human inspired trust in Car2Car communication helps to reach a better decision rate, without any central authority. The external validity of these results is limited by the assumptions of the self-awareness of capabilities, honest reputation and the boolean reality. Still this gives a strong indication of the Potential of peer trust concepts. The thesis focuses on the evolution of the trust network and lays the foundations for further research on trust networks based on statistical simulations, not only in Car2Car, but also for other applications.
If you have any questions, please contact me or open a ticket in github.