A Machine Learning Approach to Predicting Open Access Support in Research Projects

In recent years, Open Science has emerged as a pivotal element in research and innovation policies at various levels, advocating for immediate and unrestricted access to publicly funded scientific research outcomes. While the benefits of open access to research are recognized, measuring its extent poses a challenge. In this paper, we present a pioneering approach for predicting the level of support for open access within research projects by applying machine learning algorithms. Leveraging various text-mining techniques to create relevant features, our methodology utilizes machine learning models to forecast the extent of open-access support. We incorporate diverse features, including openaccess publications by researchers and organizations, funding sources, research disciplines, and interdisciplinarity, to enhance predictive capabilities. The evaluation of our proposed approach involves implementing machine learning models on features extracted from projects available on the FRIS portal. Our findings unveil a significant correlation between the degree of open-access publications of researchers and the degree of open access within the associated projects. Furthermore, the selected models achieved an accuracy range of 76-85%, showcasing their effectiveness in precise predictions. The highest accuracy, 85%, was attained with Random Forest Regression, and classical machine learning algorithms outperformed deep learning counterparts.