Dynamic Fuzzy Measures Based on Neural Networks

Fusing of multiple classifiers is a way to build classification system with high performance. The fuzzy integral has received many attentions because it can express the interaction among classifiers. It is a key to determine the fuzzy measure in multiple classifier fusion system based on fuzzy integrals. To make good use of the high performance of individual classifier in local sample space, this paper proposes a dynamic fuzzy measure based on neural network. The fuzzy measure can change with different sample. The different importance of classifier and the interaction among classifiers can be expressed in time. The classification performance can be improved. Our experiment results show that this method is effective and feasible.