A Probabilistic-Based Trust Evaluation Model Using Hidden Markov Models and Bonus Malus Systems

In the paper, the uncertainty of trust is transformed into a probability vector denoting the probability distribution over possible trust states that are hidden from observation but determined by an entity´s expected performance. We suggest the use of Hidden Markov Models (HMMs) for estimating the unknown probability distributions in peer-to-peer interactions. HMMs allow us to explicitly consider an entity´s unobserved trustworthiness that influences it´s occurrences of behavioral patterns. The proposed hidden Markov processes are associated with a specified Bonus-Malus System (BMS) that is interpreted as a Markov chain with constant transition matrix and is used to simplify the structure of model and to reduce the computational complexity of parameter estimations in HMMs. The maximum likelihood estimators of the unknown HMM parameters are obtained using EM algorithm. An application of the model in the scenario of detection of probabilistic packet-drop attack has been investigated. The simulations demonstrate that the approach is capable of accurately estimating the (hidden) trust states probability distribution as well as the expected performance for the entities that have different observed behavioral patterns.