Predicting Potential Co-Authorship Using Random Forest: Case of Scientific Publications in Indonesian Institute of Sciences

Abstract

Research collaboration is one of the strength in research management due to its advantages in quantity and quality of the research. Co-authorship network is one of the proxies to evaluate the emerging research collaborations. Co-authorship that happens for the first time among a pair of author plays an important role as the key of success for their co-authorship in the future. Therefore, the research aims to build a model predicting new co-authorship as potential co-authorship. This research used scientific articles in Indonesian biodiversity research published in Scopus during 2006-2015. New co-authorship of between 4,628 pair of authors were analyzed in terms of their similarity in co-authorship network, research interest, and community to predict whether a pair of author will have a new co-authorship in future. Random forest classifier was used to build the model after applying 10-fold cross validation in various parameter and random undersampling technique as preprocessing procedures. The result shows that the similarity in network, community network, and research interest and becomes good features to predict the potential co-authorship among a pair of author. Furthermore, paired authors that predicted to be co-authored and involving authors from Indonesian Institute of Sciences are identified as the potential patners recommended for development of research teams.