A Machine Learning Model for Predicting Collaboration in Regional Co-Authorship Network

Scientific Research collaboration is one of the strengths in the research ecosystem due to its advantages in productivity and citation. Co-authorship network is one of the methods to analyze and evaluate the emerging research collaborations. Collaboration between pair of authors for the first time plays a vital role as the key to success for their collaboration in future. In this context, a focus on SAARC is highly justified, as fostering intra-regional scientific collaboration could help address shared challenges such as public health, climate change, and sustainable development, which demand collective scientific expertise. Therefore, the objective of this paper is to build a machine learning model for predicting new potential authors within South Asian Association for Regional Cooperation (SAARC) region who never collaborated for the last 20 years (2001-2020) using data from Web of Science (WoS). The co-authorship network was analyzed between two authors using structural and semantic similarities to predict whether the collaboration will happen in future or not. A proposed Meta-Learner Binary Classifier model is applied to the link prediction predictors after data pre-processing. The result shows structural and semantic features are good features to predict potential collaborators with 0.87 AUC before sampling and 0.99 AUC after sampling.