Enhancing NTMA with Simultaneous Multi-QoS Parameter Prediction using Transformer-Based Deep Learning

The increasing use of digital communication necessitates efficient and reliable computer networks. Network Quality of Service (QoS) directly influences user satisfaction, application performance, and overall business success. Network Traffic Monitoring and Analysis (NTMA) plays a critical role in network management by providing valuable insights into network behavior and enabling proactive optimization. While extensive research focuses on network traffic prediction, simultaneous prediction of multiple critical QoS parameters remains an under-explored area. This work proposes a Transformer-based deep learning model for simultaneous prediction of download speed, upload speed, ping latency, packet loss, and network congestion. The model leverages a twostage architecture: the Extract Feature Block (EFB) captures temporal features and extracts hidden patterns from historical data, while the Predict Block (PB) utilizes these features to predict future QoS values. The proposed model offers several advantages, including reduced monitoring overhead, improved prediction accuracy through interdependency modeling, proactive network optimization, and simplified model maintenance. The proposed model was tested and compared on a network traffic dataset with three popular deep learning models for time series prediction: LSTM, CNN, and CNN_BiLSTM. The results demonstrate a significant improvement in accurately predicting QoS parameters with the proposed model.