Collusion attacks remain a major problem of reputation and trust models, in mobile ad hoc networks. By covering up malicious behaviour of one another from the remaining part of the network, two or more malicious nodes may collaborate to cause damage to or disrupt the network. A number of models exist, which have been proposed to address this issue. Despite these however, the assurance of trusted communication still remains a challenge in these networks. We present a dynamic trust model that detects malicious behaviour at runtime and prevents collusion attacks. Our proposed model employs a novel approach that has the advantage of predicting the future trustworthiness of nodes, based on historical and online behaviour of nodes. This is achieved by an architecture that applies the paradigm of Dynamic Data Driven Application Systems, in solving the problem of collusion attacks in mobile networks.