ML for Network Traffic Prediction and Congestion Control

Data Communication Theory · Group presentation

Machine Learning for Network Traffic Prediction and Congestion Control

Presented by

Shahriar Abdullah Abid

ID 3125

AKM Tasdikul Islam

ID 4047

Saad Bin Sultan

ID 4077

Md. Tahsin Habib

ID 4135

Introduction to the Topic

01Users and devices

02Network traffic

03Possible congestion

04ML-assisted prediction

The problem

Many users share a network
Demand changes during the day
Heavy traffic can cause congestion, delay and packet loss

The idea

ML learns patterns from past data
It may estimate future traffic or early congestion
Predictions can support decisions

Presentation Overview

Network traffic and congestion

How machine learning makes predictions

Applications and a practical example

Congestion control, benefits and limitations

Motivation: Why Does Wi-Fi Become Slow?

Morning: few users

same network link

24%

Few users share the link
Demand is low
Network may run smoothly

During class: many users

same network link

96%

Many students download or stream
More traffic, same limited capacity
Network may become congested

Can we predict when a network will become busy before congestion occurs?

Understanding Network Traffic and Congestion

Network traffic: data travelling through a network.

Network congestion: demand exceeds the capacity available to carry it smoothly.

Low traffic

plenty of room

Congested

demand > capacity

extra packets must wait or may be dropped

Bandwidth vs. Throughput

Bandwidth

100Mbps

The most the link can carry

Throughput

40Mbps

What you actually get right now

Link capacity

Actual data flow

What Happens When a Network Gets Congested?

Demand increases

Packets wait in a queue

Queue grows: more delay

Queue becomes full

Some packets are dropped

Delay: a packet takes longer to reach its destination.

Packet loss: a packet fails to reach its destination.

What Is Machine Learning?

Historical data

Records from past days

Learn patterns

Find statistical patterns

Make a prediction

Estimate a new situation

Wi-Fi example: past days show heavy traffic around 10 am, so the model can predict a busy period.

How Does ML Learn from Network Data?

Data

information collected from the network

Features

the details used to make a prediction

Training

learning patterns from past examples

Prediction

an estimate of what may happen next

Illustrative features

How Can ML Predict Network Traffic?

Records from previous days

Model learns: traffic often rises around 10 AM

It uses patterns and inputs to estimate future traffic

Staff may use the estimate to prepare

Traffic level

Illustrative

From Traffic Prediction to Congestion Warning

Traffic prediction: how much traffic may arrive.

Congestion prediction: will capacity be insufficient?

Heavy traffic, enough capacity

No congestion

capacity

demand

Heavy traffic, limited capacity

Congestion

capacity

demand

Measure

Estimate

Assess risk

Possible warning

The Role of ML in Network Management

ML predicts. People or network mechanisms decide what to do.

Prediction

Network measurements

ML model learns from past data

Prediction: traffic or congestion risk

Control

People or network mechanisms decide

Prepare capacity

Use another path, if available

Adjust allowed settings

Applications of ML-Assisted Prediction

Campus Wi-Fi

Predict busy times so staff may prepare.

Video calls and streaming

Traffic estimates may help services adapt to changing conditions.

Mobile networks, including 5G

Predictions may support planning for high-demand periods.

Cloud and data centers

Predictions may help identify links likely to overload.

Practical Example: Campus Wi-Fi

Conventional

Illustrative example · not real data

Preparation comes late: congestion, delay or packet loss may follow.

Possible ML-assisted

More time to prepare, e.g. spread users across access points. Congestion can still happen if capacity is too small.

Methodology 1: Conventional Congestion Control

Networks already respond to congestion using established rules and protocols.

TCP congestion control

TCP, used by most apps, makes the sender slow down when it sees congestion signs. Versions differ.

Network management

Capacity planning: add capacity early

Prioritization: key traffic goes first

Routing: choosing the path packets take

Methodology 2: Proposed ML-Assisted Workflow

ML prediction

Decision and follow-up

Collect data

traffic, delay, packet loss

Learn patterns

from historical examples

Predict

traffic or congestion risk

Support a decision

a warning or possible response

Monitor

compare with the prediction; repeat

Early warning

An alert while the queue is still filling, before packets are dropped. It helps only if the prediction is accurate enough and the network can act.

Comparison: Conventional vs. ML-Assisted

Traffic

Time

Capacity

Traffic demand

ML predicts

Earlier warning

Congestion

Conventional responds

Basis

Proven rules and protocols, using measurements

Learned patterns plus current measurements

Strength

Well tested; adapts to changing conditions

May anticipate traffic or congestion earlier

Limitation

Rules may not fit every situation well

Needs good data; predictions may be wrong

ML may complement existing congestion control; it does not replace it.

Why Network Congestion Matters

Data communication aims to transfer data between devices effectively and reliably.

Capacity: what the link can carry.

Demand: what users want to send.

Delay can increase

Packets may be lost

Throughput can fall

Benefits and Limitations

Possible benefits

Earlier warning

More time to prepare for expected traffic

Potentially better use of network resources

Limitations

Predictions may be wrong

Traffic patterns can change

Good data is needed

Testing and computing resources are needed

Existing congestion control remains important

Conclusion

Machine learning may help networks anticipate traffic and congestion, but it complements existing congestion-control mechanisms rather than replacing them.

Congestion can increase delay and cause packet loss.

ML can learn patterns from historical network data and make predictions.

The usefulness of predictions depends on data quality, accuracy and the network’s ability to respond.

References

Cloudflare. How to Make the Internet Faster for Everyone. Cloudflare Learning Center. Accessed October 2026.

IBM. What Is Machine Learning? IBM Think. Accessed October 2026.

Cisco. QoS Frequently Asked Questions. Cisco Support, document 22833, updated 4 June 2009. Accessed October 2026.

Cisco. QoS Congestion Avoidance. Cisco technology page. Accessed October 2026.

Allman, M., Paxson, V. and Blanton, E. TCP Congestion Control. RFC 5681, IETF, September 2009.

Thank You