Research on the Construction of Communication Network Traffic Prediction Model Based on Machine Learning

A multi-model prediction architecture based on the integration of LSTM, XGBoost and SVR is constructed for the complex timing characteristics of communication network traffic. The model realizes accurate prediction of short-term network traffic by combining weighted fusion strategy based on sliding time window and multi-dimensional feature engineering. The experimental results show that the LSTM model has an RMSE of 12.41 on the test set, which is significantly better than XGBoost's 14.07 and SVR's 15.64, and the linear regression error is as high as 18.29, which is the weakest performance. The fusion model further improves the stability and performs better in terms of error distribution and anomaly control, providing effective support for the optimal allocation of communication network resources.