Training Data Selection for Record Linkage Classification

This paper presents a new two-step approach for record linkage, focusing on the creation of high-quality training data in first step. The employs unsupervised random forest model as similarity measure to produce score vector matching. Three constructions were proposed select non-match pairs data, with both balanced (symmetry) and imbalanced (asymmetry) distributions tested. top construction was found be most effective producing 100% correct labels. Random support machine classification algorithms compared, produced an F1-score comparable probabilistic linkage using expectation maximisation algorithm EpiLink. On average, forests improved by 1% recall 6.45% compared existing methods. By emphasising this has potential improve accuracy efficiency wide range applications.