Trustworthiness prediction of cloud services based on selective neural network ensemble learning

Cloud services have become a popular and flexible solution for providing components to build service-based systems. A component's trustworthiness is a key measure that can guide service requesters when making a service selection decision. Prediction of this trustworthiness, based on the component's multi-faceted quality of service (QoS) attributes, is therefore an important problem to address. In this paper, selective ensemble learning is introduced to address the trust problem for cloud services: We use back-propagation neural networks (BPNNs) as the basic classifiers, with two swarm intelligence algorithms adapted to search for the optimal aggregation weights to create the ensemble: Basic particle swarm optimization (PSO) is used for decimal weights; and quantum discrete PSO (QPSO) is used for binary (0-1) weights. The optimized ensemble learning model, based on BPNNs, is then used to predict the trustworthiness of a given cloud service. Extensive experiments are performed on a well-known, public dataset to verify the effectiveness of the proposed trust prediction algorithms. The experimental results show that our algorithms are not only better than the basic BPNN method in prediction precision, but also outperform current state-of-the-art trust prediction algorithms. The proposed algorithms also exhibit a strong robustness.