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. This paper presents a novel machine learning-based approach to assess and predict the level of open-access support within research projects. By analyzing key indicators such as publication practices, funding sources, research disciplines, and interdisciplinarity, we develop predictive models that identify patterns and trends in open-access adoption. Our analysis, applied to Flemish research projects from the FRIS portal, reveals strong links between open-access publication activity and the broader adoption of open-access principles in research projects. The models demonstrated high predictive accuracy, with results highlighting the effectiveness of classical machine learning techniques over deep learning approaches. In addition, we explored advanced machine learning techniques such as ChatGPT and LLaMA, which offer significant improvements in natural language processing and predictive capabilities. These techniques enhance our ability to analyze complex datasets and generate more accurate predictions regarding open-access support. Future work will focus on extending our predictive models to other research outcomes. This will enable a more comprehensive understanding of open-access support across various dimensions of scientific research. This work provides actionable insights for researchers, data stewards, policymakers, funders, and project managers, supporting evidence-based strategies to foster and evaluate open-access practices, thereby advancing the goals of Open Science.