Network traffic analysis and bandwidth forecasting for using Meta’s Prophet: a case study

Isaac, Yusuf Onimisi; Bamisaye, Ayodeji James; Adedotun, Ijagbemi; Dada, Theophilus Olusegun; John, Onyemenam Obiajulu · 2026 · Crossref

DOI: 10.12928/telkomnika.v24i3.27609

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the challenge of network congestion and quality of service (QoS) degradation in high-density academic environments by developing a bandwidth forecasting system for student halls of residence at Landmark University. Motivated by the irregular and bursty nature of modern internet traffic—driven by activities like video streaming and cloud backups—the research aims to provide network administrators with predictive tools to optimize resource allocation and prevent outages. The authors selected Meta’s Prophet, an open-source time-series forecasting model, due to its interpretability, ability to handle multi-seasonality, and lower computational requirements compared to deep learning alternatives like LSTM or complex hybrid models. The methodology employed a quantitative case study design using historical network traffic data collected from October 1 to December 15, 2024. Data was gathered hourly from eight residence halls using Paessler Router Traffic Grapher (PRTG) monitoring software, capturing metrics such as bandwidth usage, latency, and uptime. The raw data underwent rigorous preprocessing, including conversion to CSV format, handling of missing values via seasonal interpolation, and outlier removal using the interquartile range method. The dataset was split into 80% for training and 20% for testing, preserving temporal order to simulate real-world forecasting scenarios. The Prophet model was configured to account for daily and weekly seasonality, trend changes, and holiday effects, with performance evaluated using mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE). The results demonstrated that the Prophet model achieved over 90% accuracy in predicting future bandwidth usage. The calculated MAE was 10,099,863.10 bits per second (bps), and the RMSE was 13,570,959.58 bps. While the mean squared error (MSE) appeared numerically large, the authors contextualized these errors as reasonable relative to the observed peak bandwidth usage, which fluctuated between 47 and 50 megabits per second (Mbps). The model successfully captured diurnal seasonality, accurately predicting usage peaks during evening hours (7 PM–11 PM) and troughs during early morning hours. However, the model exhibited minor limitations, including a slight phase lag during sharp traffic transitions and underestimation during rapid declines in usage, likely due to its smoothing mechanisms. The study concludes that Meta’s Prophet is a viable and efficient tool for short-term bandwidth forecasting in campus networks with predictable usage cycles. The findings suggest that machine learning-based forecasting can significantly aid in proactive network management, capacity planning, and anomaly detection. The authors recommend that while Prophet offers a strong balance of accuracy and resource efficiency, it may need to be complemented by anomaly detection systems or hybrid models in environments characterized by high volatility or frequent unpredictable spikes. This work provides a scalable framework for improving network infrastructure and user experience in similar institutional settings.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• telematics crash prediction