Reliability evaluation of two-stage evidence classification system considering preference and error

Abstract

In classic weighted voting systems (WVSs) and weighted voting classifiers (WVCs), the uncertainty in the system is characterized by probability theory, and the decisions of voting units are fused according to the majority rule. We use D-S evidence theory to improve the representation of the voting unit's decision and the fusion rule in WVSs and WVCs, and propose a new evidence classification system (ECS). In the ECS, a voting unit's decision is represented as a basic probability assignment function, the majority rule is replaced by the combination rule in D-S evidence theory. We also extend the WVS or WVC's decision-making process into two stages, and consider the measurement error and decision preference of the voting unit. A method based on Monte Carlo simulation is proposed to evaluate the reliability of the ECS. The effects of the number of voting units, decision preference and measurement error on the system reliability are also analyzed. Simulation results show that the introduction of evidence theory into ECS can effectively eliminate the effect of preference and error on system reliability.