Research on hybrid federated learning algorithm integrating dynamic contribution degree

Aiming at the problems of data heterogeneity among participants and poor communication efficiency in traditional federated learning, this study proposes a Dynamic Contribution-aware Hybrid Federated Learning (DCHFL) algorithm. In the DCHFL algorithm, hybrid federated learning integrates horizontal federated learning and vertical federated learning to address the issue of data heterogeneity; the integration of dynamic contribution degree is intended to solve the problem of communication efficiency. For hybrid federated learning, based on the feature complementarity of horizontal and vertical federated learning, a hierarchical parameter aggregation mechanism is designed to realize the collaborative training of sample space and feature space. For dynamic contribution degree, considering the differences in data distribution among participants, the quality of local model updates, and communication resource constraints, a dynamic contribution degree evaluation strategy based on the Shapley value is introduced to quantify the real-time contribution weight of the global model and adaptively optimize the communication frequency and participant selection strategy. Experimental results show that the prediction accuracy of the DCHFL algorithm is 0.9992, which is superior to the five comparison methods in the literature, further verifying the effectiveness and feasibility of the proposed algorithm. The contribution of this study lies in providing theoretical support and practical solutions for the communication optimization of federated learning in dynamic heterogeneous environments.