Scientific research is expanding rapidly, producing vast volumes of scholarly literature across diverse disciplines. However, identifying emerging trends and uncovering interdisciplinary intersections remains a persistent challenge due to the fragmented and nonlinear nature of modern knowledge production. Traditional bibliometric methods such as co-citation analysis and keyword co-occurrence rely on pairwise relationships and often fail to capture the higher-order associations that drive innovation. In this study, we present a novel hypergraph-based framework for forecasting scientific research trends. In our approach, nodes represent research concepts, and hyperedges correspond to publications that link multiple concepts. By framing the task as a hyperedge link prediction problem, we uncover latent conceptual groupings that may signal future research directions. We then develop hypergraph neural network models, HLP, HyperGCN, and HyperSAGE, alongside traditional graph-based models, using a real-world dataset of scientific concepts used in papers from the field of Ethnic Studies. Our results show that hypergraph models consistently outperform their graph-based counterparts in accuracy and predictive power.