Consumer Price Index Forecasting Based on Federated Learning and a Hybrid Deep Neural Network for Economic Activity Evaluation

ABSTRACT Transformer‐based model has achieved significant success for consumer price index (CPI) prediction task. Its higher model complexity and limited local modeling ability require more training samples. However, privacy protection in different economic sectors severely restricts predictive performance. Although distributed learning can achieve cross‐institutional collaborative training, resource‐constrained clients pose significant challenges for the deployment of deep models. To this end, this paper proposes a 1D convolution and Swin Transformer hybrid network based on federated learning for CPI prediction, in which the 1D convolution operations focus on local features and the Swin Transformer achieves long‐distance dependency modeling with lower computational complexity. In addition, we introduce federated learning mechanisms to further improve model training in privacy protection scenarios. The experimental results on the public dataset indicate that the proposed model achieves the highest predictive performance by introducing the distributed learning measures.