Predictive Analytics Models for Network Traffic Optimization in Distributed Enterprise Environments

Abstract

This study examined the effectiveness of predictive analytics models for network traffic optimization in distributed enterprise environments, with particular emphasis on forecasting accuracy, computational performance, dynamic resource allocation, congestion reduction, latency, throughput, and bandwidth utilization. A quantitative experimental design was employed using 120,000 valid network traffic observations retained from an initial dataset of 128,640 records, representing a 93.28% retention rate after preprocessing. The final dataset was chronologically divided into 70% training, 15% validation, and 15% testing subsets, with 18,000 observations reserved for independent model evaluation. Six predictive approaches—Autoregressive Integrated Moving Average (ARIMA), Support Vector Regression (SVR), Random Forest, XGBoost, Gated Recurrent Unit (GRU), and Long Short-Term Memory (LSTM)—were evaluated using MAE, MSE, RMSE, MAPE, R², training time, and inference time. LSTM demonstrated the strongest forecasting performance, achieving an MAE of 21.84 Mbps, RMSE of 28.50 Mbps, MAPE of 3.91%, and R² of 0.958, followed by GRU with an RMSE of 30.42 Mbps and R² of 0.952. Relative to ARIMA, LSTM reduced MAE by 43.21% and RMSE by 43.94%. Predictive traffic management also produced substantial operational improvements compared with non-predictive management. Bandwidth utilization increased from 67.84% to 76.92%, throughput increased by 16.09%, and resource-allocation efficiency improved by 18.09%. End-to-end latency decreased by 24.70%, packet loss declined by 36.36%, congestion frequency decreased by 37.05%, and congestion duration was reduced by 34.07%. Prediction error was significantly associated with congestion frequency (r = 0.61, p < 0.001) and negatively associated with resource-allocation efficiency (r = −0.63, p < 0.001). Significant effects were also identified for predictive model type, traffic-load intensity, and forecasting horizon. The findings demonstrated that predictive analytics, particularly recurrent deep learning models, provided measurable improvements in traffic forecasting and network optimization while highlighting the importance of balancing predictive accuracy, computational efficiency, workload intensity, and forecasting horizon in distributed enterprise network management.