A novel ensemble learning approach to unsupervised record linkage

Highlights：

• A novel unsupervised approach to record linkage has been proposed.

• The approach combines ensemble learning and automatic self learning.

• An ensemble of diverse self learning models is generated through application of different string similarity metrics schemes.

• Application of ensemble learning alleviates the problem of having to select the most suitable similarity metric scheme and improves the performance of an individual self learning model.

• The proposed method obtained comparable results with the supervised methods.