An effective multi-classifier strategy for attributes matching in heterogeneous databases

Abstract

Attributes matching is crucial to implement data sharing and interoperability across heterogeneous databases. In order to lay a theoretical foundation for our multi-classifier strategy, we analyze the limitations that use a single trained neural network to determine the corresponding attributes. The analyzed result demonstrates the theoretical possibility that different input vectors map the same output vector by a single-trained neural network. This could result in mismatching and consequently decrease the matching accuracy. Based on the theoretical result and instance analysis, an effective multi-classifier algorithm for attributes matching is proposed in this paper. We verify the correctness of our analysis and proposed multi-classifier strategy by using our previous experimental results.