<p>Addressing the problem of uneven curriculum resource allocation in interdisciplinary integration courses in higher education and the difficulty of effectively identifying independent resource attributes, this study proposes a reinforcement learning-based curriculum resource allocation method. Unlike traditional static allocation approaches, the proposed framework integrates curriculum resource feature selection, resource classification, and reinforcement learning-based decision optimization into a unified allocation mechanism capable of adapting to dynamic educational demands. First, a teaching model for interdisciplinary integration courses is developed by incorporating multidimensional resources, including faculty, laboratory equipment, funding, and time, to establish a comprehensive digital curriculum resource database. Subsequently, the XGBoost algorithm is employed to perform feature selection, and the selected features, together with the original curriculum resource data, are used to construct a multi-label curriculum resource classification model based on decision trees for resource category identification. A multi-objective resource allocation model is then established to achieve global balance in multi-attribute resource allocation, dynamically optimize allocation strategies, and improve the quality of allocation outcomes. To solve the proposed model, a Twin Delayed Deep Deterministic Policy Gradient (TD3) reinforcement learning algorithm incorporating behavior cloning is adopted for offline training and online optimization. The action space includes dynamic resource scheduling, curriculum adjustment, and budget allocation, while the state space is defined using resource status, curriculum requirements, historical allocation information, and external environmental factors. A reward function is designed to guide policy optimization and generate real-time curriculum resource allocation strategies. Experimental results demonstrate that the proposed method achieves an absolute matching rate of 97.59% in curriculum resource classification while reducing computation time by 71% during feature screening. Furthermore, the utilization rate of curriculum resources remains between 95 and 98% after allocation, indicating high allocation efficiency, strong operational stability, and a substantial reduction in resource idleness and waste. These results demonstrate the practical value of the proposed framework for supporting intelligent resource management and improving decision-making efficiency in interdisciplinary higher education environments.</p>