An alternative trust calculation method using radial basis function neural networks

Abstract

One of the most challenging problems in an electronic environment is the trust between electronic entities. Trust is a generic concept and can be effectively used in various contexts. An important question is: “How does an electronic entity trust another electronic entity?” or “How can trust be determined?” The most important challenges identified in trust calculation are increased calculation complexity, data storage and access to large data sets for trust calculation. Most of the trust calculations are in the security context, and each electronic entity has its features and standards. A radial basis function neural network (RBFNN) is a feed-forward neural network used as a universal function approximator to solve nonlinear problems. Training an RBFNN to model trust values requires accurate, large training data sets. Insufficient trust data sets were found due to privacy and data storage problems. The increased calculation complexity and the data storage problem may be solved using an RBFNN. In addition, a possible solution to trust data scarcity is synthetic data generation. The primary purpose of this study is to find an alternative trust calculation method using RBFNNs. To achieve this, literature regarding different forms of trust calculation is considered, including identifying the three dimensions of trust that leads to a new definition for the term trust. The three dimensions of trust specified are the trust context, calculations for the quantification of trust, and information sources. Challenges in calculating trust in an electronic environment are investigated. Obstacles in using an RBFNN for the calculation of trust are identified. Literature regarding the creation of synthetic trust data is reviewed, and the challenges in generating synthetic trust data are addressed by a seven-step framework called the PSTDG (Pure synthetic trust data set generation) Framework. Consequently, a new definition for the term validation of generated trust data is developed. In addition, a three-step method is created to validate the data generation model. Finally, a four-step experimental design process is developed to build a model using an RBFNN that can determine trust values between electronic entities. The four-step experimental design process was demonstrated by performing two experiments. In the first demonstration, a data generation model called the PSTDG-PeerTrust model based on PeerTrust, a purely theoretical trust calculation model, is developed. The PSTDG-PeerTrust model is validated, and a trust data set is generated. An RBFNN model called PeerTrustRBFNN is then built, using the best model hyperparameters found. In the second demonstration, the Amazon Relational Database Service (ARDS) shows that real-life problems can also be solved

using the proposed method. Hence, a data generation model called PSTDG-ARDS is developed. This model is validated, and a trust data set is generated. The ARDSTrustRBFNN trust model is built, using the identified best model hyperparameters. The study shows that the data generation models developed using the PSTDG Framework can produce valid pure synthetic trust data, and an RBFNN can be used to calculate trust in an electronic environment.