A Comparative Study of Multi-Label Classification for Document Labeling in Ethical Protocol Review

Abstract

An ethical clearance document ensures that the research will protect the subject in accordance with existing ethical principles. The ethical clearance is issued by the Research Ethics Commission (KEP). KEP will conduct a review of the proposed ethical protocol based on the seven standards contained in a protocol. The review process is done manually by KEP. This process often creates bottlenecks in research due to the large number of protocols that must be reviewed, so that the process to get ethical clearance takes a long time. This can affect the setback in the schedule of the research process. Therefore, in this research, a comparative study was conducted on the problem of multi-label classification to automate the ethical protocol review process. Automation of the labeling process can increase the effectiveness of the review process because it can provide an overview to the reviewer regarding the label of a document before conducting a more in-depth review process. The experiment results show that the use of the traditional machine learning approach produces better performance than the deep learning approach. The machine learning method with the best results is Naïve Bayes+BoW with precision, recall, and F-score values of 0.76, 0.80, and 0.78, respectively.