This repo shows a step-by-step process to download and audit Pubmed data and then use it to create a deep-learning case-report classifier. This classifier might be used to identify unsuitable submissions for journals which don't consider case-reports.
Equally, the same code could be adapted to create any classifier, such as a classifier for another article-type, or a particular subject area.
The best results come from either a support-vector classifier, or the neural-network approach. An ensemble model may yield better results.