A Hybrid Network Congestion Prediction Method Integrating Association Rules and LSTM for Enhanced Spatiotemporal Forecasting

Abstract

With the acceleration of the Internet of Everything, network congestion has become increasingly serious, affecting network operation efficiency and user experience. Therefore, accurate prediction of network and congestion conditions is of great significance for optimizing network management systems. This study proposes a network congestion prediction method based on association rules and a long short-term memory network (LSTM) to improve the accuracy of network prediction. First, we use the association rule algorithm to mine the potential relationship between different time periods, gateway networks, and environmental factors, extract highly relevant features from historical network data, and use them as part of the input features of the LSTM model. Subsequently, LSTM further learns time series patterns by modeling the time dependency of the network to achieve accurate prediction of future network conditions. The experiment uses the PeMS network dataset for verification and compares it with a variety of benchmark models (ARIMA, SVR, CNN-LSTM, Transformer, etc.). The experimental results show that the proposed method outperforms other methods in terms of MSE, RMSE, and MAPE indicators, especially in complex network scenarios, and has stronger prediction stability. In addition, to further analyze the effectiveness of the model, we conducted an ablation experiment. The results show that association rules can effectively improve the prediction ability of LSTM, and feature engineering also has a significant impact on the accuracy of the model. This study also analyzed the effects of different optimizers (SGD, AdamW, Adam) and learning rates and found that the Adam optimizer and a smaller learning rate (0.001) can improve the convergence stability and prediction accuracy of the model. Although this study has achieved good results in network prediction, there are still some challenges, such as the joint prediction of multiple gateways and real-time data fusion. Future research can explore new deep learning architectures, such as Transformer and graph neural network (GNN), to further optimize network prediction and combine reinforcement learning to improve the adaptability of the model, providing stronger technical support for the development of network management systems.