Scholarlink

<p>Accurate traffic forecasting is crucial for efficient resource management in 5G networks. Traditional RNN models struggle to handle complex patterns and large-scale data in such networks. This study introduces an Auto-encoder decoder-based LSTM model for Traffic Forecasting in 5G Networks using Federated Learning. The Federated Learning setup of the model allows decentralized training, ensuring privacy by keeping raw data local while a central server aggregates model updates. Experimental results show that the proposed model outperforms traditional RNNs in forecasting accuracy and capturing complex dependencies effectively. The empirical evaluation demonstrates that the proposed architecture exhibits statistically significant performance improvements over existing baseline models across multiple evaluation metrics. Experimental results demonstrate that the proposed Dual-Attention Autoencoder-Decoder LSTM architecture, integrated with FedAvgMomentum, achieves a significant improvement over baseline models. Specifically, our model reduces the Mean Absolute Error (MAE) by 14.6%, the Root Mean Square Error (RMSE) by 12.3%, and the Normalized RMSE (NRMSE) by 11.8% compared to conventional LSTM-based federated models on real-world 5G LTE datasets. The analysis shows substantially lower error rates in key performance indicators.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Privacy-Preserving 5G Traffic Prediction Using Federated Learning with Dual-Attention Mechanisms

  • 期刊论文

  • 发表日期：

    2026年4月9日

查看全文
  • Abdul Kalam,
  • Samayveer Singh,
  • Aruna Malik,
  • Aviral Sangal

摘要

Accurate traffic forecasting is crucial for efficient resource management in 5G networks. Traditional RNN models struggle to handle complex patterns and large-scale data in such networks. This study introduces an Auto-encoder decoder-based LSTM model for Traffic Forecasting in 5G Networks using Federated Learning. The Federated Learning setup of the model allows decentralized training, ensuring privacy by keeping raw data local while a central server aggregates model updates. Experimental results show that the proposed model outperforms traditional RNNs in forecasting accuracy and capturing complex dependencies effectively. The empirical evaluation demonstrates that the proposed architecture exhibits statistically significant performance improvements over existing baseline models across multiple evaluation metrics. Experimental results demonstrate that the proposed Dual-Attention Autoencoder-Decoder LSTM architecture, integrated with FedAvgMomentum, achieves a significant improvement over baseline models. Specifically, our model reduces the Mean Absolute Error (MAE) by 14.6%, the Root Mean Square Error (RMSE) by 12.3%, and the Normalized RMSE (NRMSE) by 11.8% compared to conventional LSTM-based federated models on real-world 5G LTE datasets. The analysis shows substantially lower error rates in key performance indicators.