An ensemble model for collective classification that reduces learning and inference variance
Year: 2012
Type: article
Abstract: Ensemble learning can improve classification of relational data.Previous attempts to do so include methods that have focused primarily on reducing learning or inference variance, but not both at the same time.We present an ensemble model that reduces error due to variance in both learning and collective inference.Our model uniquely combines two strategies tailored specifically for relational data and relational models to achieve a larger reduction in variance than using either method alone, which results in significant accuracy gains.In addition, we present the first theoretical analysis for ensembles of collective classifiers in relational domains, to show the reasons for the superior performance of our proposed method.We also use synthetic and real world data to demonstrate the improvement empirically.
Authors: Hoda Eldardiry, Jennifer Neville
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W1603982372