Screening PubMed abstracts: is class imbalance always a challenge to machine learning?

Resampling techniques slightly improved the performance of the investigated machine learning techniques. From a computational perspective, random undersampling 35:65 may be preferred.