Reinforcement Learning-Controlled Subspace Ensemble Sampling for Complex Data Structures

This paper addresses the challenges of structural complexity and sample bias in high-dimensional imbalanced data mining by proposing a reinforcement learning-controlled subspace ensemble sampling adaptive mining algorithm. The method integrates a policy-driven subspace selection mechanism with a dynamic ensemble sampling module to jointly optimize feature dimensions and sample distribution. In the subspace construction phase, a reinforcement learning agent adjusts the feature selection strategy dynamically based on classification performance feedback. During the sampling phase, the method combines under-sampling and over-sampling strategies to guide sample redistribution, effectively alleviating performance bottlenecks caused by class imbalance. A series of comparative experiments are designed to evaluate the algorithm's performance under different embedding dimensions, generalization requirements, and sampling sensitivity. The results show that the proposed method consistently outperforms traditional approaches in terms of Accuracy, F1-score, and other core metrics. It demonstrates stronger stability and adaptability in complex data structures. This study provides a new strategy modeling approach for high-dimensional data mining. It contributes significantly to enhancing the robustness of imbalanced learning algorithms in real-world applications.