Connectionist based Dempster-Shafer evidential reasoning for data fusion

A network realization of the Dempster-Shafer evidential reasoning is developed, and it is further extended to a neural network, referred to as DSETNN, for dealing with the dependence of evidential sources. DSETNN is tuned for optimal performance through a supervised learning process. To demonstrate the effectiveness of DSETNN, we apply it to two benchmark pattern classification problems. Experiments reveal that DSETNN outperforms the Dempster-Shafer evidential reasoning, the majority voting, single source based results, and provides encouraging results in terms of classification accuracy and the speed of learning convergence.