5G Traffic Prediction with Time Series Analysis

Nikhil Nayak, Rujula Singh R

Abstract: In today's day and age, a mobile phone has become a basic requirement needed for anyone to thrive. With the cellular traffic demand increasing so dramatically, it is now necessary to accurately predict the user traffic in cellular networks, so as to improve the performance in terms of resource allocation and utilisation. By leveraging the power of machine learning and identifying its usefulness in the field of cellular networks we try to achieve three main objectives: classification of the application generating the traffic, prediction of packet arrival intensity, and burst occurrence. The design of the prediction and classification system is done using a Long Short Term Memory (LSTM) model. The LSTM predictor developed in this experiment returns the number of uplink packets and estimates the probability of burst occurrence in specified future time intervals. For classification, the regression layer in our LSTM prediction model is replaced by a softmax classifier to categorize traffic into surfing, video calling, voice calling, and video streaming.

Introduction

As 5G networks expand, the demand for data is skyrocketing, leading to highly "bursty" and unpredictable traffic patterns. Traditional network management often struggles to allocate resources like bandwidth and power dynamically because it cannot accurately foresee these sudden spikes. A specific technical hurdle identified in this research is the sparsity of network traffic data, which is often full of "zeros" when no packets are sent. Standard machine learning models frequently fall into a "zero-prediction" trap, achieving deceptively low error scores by predicting zero most of the time while failing to capture the rare but critical traffic bursts that cause network congestion.

System Architecture and Methodology

The researchers employed a two-pronged approach using Deep Learning and Unsupervised Learning to address these challenges. The end-to-end pipeline involves raw data acquisition, specialized preprocessing, and a modular neural network architecture.

End-to-end pipeline for 5G traffic analysis.

Time Series Prediction (LSTM)

An LSTM network with 100 units followed by a fully connected layer was utilized. LSTMs are specifically designed to capture long-term temporal dependencies, making them ideal for time-series data like packet flows. To combat the "zero-bias" issue, the team implemented ProPadding and a NonZeroPrediction strategy. This involved adjusting the training set to prevent the model from being overwhelmed by empty time bins, forcing it to learn the actual trends of active traffic.

Traffic Classification

To understand the nature of the traffic, the study used K-Means clustering to group traffic into four categories. For supervised classification tasks, the LSTM's regression layer was replaced with a Softmax classifier to distinguish between applications such as video streaming and voice calling.

Datasets and Feature Extraction

The study utilized a dataset of raw network traffic captured via WireShark. The raw data was processed through MATLAB to extract features including source/destination IP addresses, uplink/downlink packet counts and sizes, protocols, and timestamps. These features were aggregated into various time bins (1s, 3s, 30s, and 60s) to test the model's sensitivity to temporal resolution.

Key Results

By filtering out the influence of zero-heavy history through NonZeroPrediction, the authors significantly lowered the Root Mean Square Error (RMSE) and achieved a trend line that closely matches actual traffic spikes.

Prediction accuracy and classification performance metrics.

The final LSTM model demonstrated high precision and recall for identifying imminent traffic bursts. As shown in the classification reports, the model achieved an overall accuracy of 91%, with particularly strong performance in identifying baseline traffic (class 0.0) and reliable detection of burst events (class 1.0).

Conclusion and Future Work

This research highlights that a "good" RMSE can be deceptive in sparse network environments; predictive utility depends on capturing bursts rather than minimizing average error. While the current model successfully segments traffic and predicts intensity, future work will focus on generating real-world 5G data for better supervised validation and integrating diverse traffic patterns from Internet of Things (IoT) devices.

Summary updated

BibTeX
@article{nayak2021g,
  author = {Nikhil Nayak and Rujula Singh R},
  title = {{5G Traffic Prediction with Time Series Analysis}},
  journal = {International Journal of Innovative Technology and Exploring Engineering},
  year = {2021},
  doi = {10.35940/ijitee.l9555.10101221},
}
Figure