Blockchain Quantitative Model Training Method Integrating Image Processing and Federated Learning: Asset Price Prediction for Multi-Institution Privacy Protection

View at publisher Summarise with VIRAT
Authors: Xiaosen Zeng, Biao Tian, Haoyuan Yang, Zhaoxuan Zhang, Mengzhen Li, Yuwei Wang, Yue Chang, Yiyue Hu, Ziyun Zhang
Cited by: 0
References: 17

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Along with the rapid increase of multi-institution financial data, the traditional centralized forecast model is unable to achieve high precision asset price prediction while maintaining privacy. In order to solve this problem, we propose a new approach to improve the performance of the model, which is based on the blockchain-based approach. Through the mapping of the time-series data to the Gramian angular domain image, we use the CNN-ViT hybrid feature extraction network, which enables the encryption of parameters to be aggregated. Experiments show that the proposed approach achieves a RMSE of 0.0182 on the NASDAQ data set, which is 26.6% less than the conventional centralized training, and at the same time it improves R2to 0.958. It maintains over 93.9% model stability in multi-node environments, significantly outperforming conventional federated models. The system combines prediction precision, privacy protection, and traceability, and provides an efficient way for multi-institution financial collaborative modeling.

Fields of Study & Subjects

Stock Market Forecasting Methods Blockchain Technology Applications and Security Privacy-Preserving Technologies in Data

Publication details

Journal / source
IEEE International Conference on Power, Electronics and Green Energy (ICPEGE)
Pages
1507–1511
Publication date
28 January 2026
Language
EN