Bandwidth Usage Forecast with Machine Learning

How We Forecast Bandwidth Usage With Machine Learning

It is difficult to find an area of expertise in which intelligent systems cannot be applied. Scientific and practical solutions are carried out day by day faster than ever before using such intelligent systems. Researching and Developing Intelligent CDN solutions is the core objective of Medianova CDN R&D Team. Within a CDN perspective, intelligent systems can be applied to many specific areas such as Request-Routing, Point of Presence (PoP) Assignment and Traffic Management.

In order to incorporate automatized intelligent-decision-making processes within our CDN infrastructure; we resort to Artificial Intelligence using Machine Learning Algorithms. In this blog post, after we explain general concepts of Artificial Intelligence and Machine Learning, we will show you how we predict our Bandwidth Usages using Artificial Neural Networks.

can machines think

-Alan Turing, Can Machines Think?

What is Artificial Intelligence?

People need other people’s help in solving difficult problems, and other people need more people to solve such specific problems. For such complex problems; Artificial Intelligence becomes handy when it comes to solving problems on its own. Artificial Intelligence is the science dedicated to developing machines that are working, thinking, and operating as if like people.

We see Artificial Intelligence almost in every aspect of our lives when we use our phones, computers, cameras etc. Few examples are auto-correction and text recommendation; face detection and recognition; intelligent personal assistants like Siri, Alexa.

We utilize Artificial Intelligence when it comes to predicting Bandwidth Usage in our servers, so we can reduce our costs and automatize CDN traffic.

What is Machine Learning?

Machine Learning is a branch of AI which is a general term for “trainable algorithms”. “Trainable“ refers to an algorithm that analyzes huge amounts of data, and then updates its own parameters according to analyses. OK, so why do we need a trainable algorithm? Unless you’re using Machine Learning then you define all your conditions statically in traditional programming; on the contrary, trainable algorithms updates these conditions depending on the data feed. For instance; if you want to predict your bandwidth usage with traditional programming, you have to define that

Screen Shot 2020-05-12 at 3.22.44 PM

You could be sure that there are thousands of such conditions that you can write. In Machine Learning you only need to feed data to the trainable algorithm correctly, then the algorithm is trained and makes the correct decisions.

In Medianova CDN, we use Artificial Neural Networks as our trainable algorithm to forecast Bandwidth Usage.

What is an Artificial Neural Network?

The Artificial Neural Network (ANN) is one of the machine learning algorithms created inspired by the neural mechanics of the human brain. ANN has input-, hidden-, and output-layers; where each layer consists of neurons that are essentially trainable algorithms. Artificial Neural Networks are widely used in studies focused on advanced artificial intelligence studies such as Deep Learning.

Artificial Neural Network Model

Why Do We Use Artificial Intelligence?

Data is the future. With the advances in machine learning, data collection has become more important than algorithms. All companies have their specific sets of data; however, they do not utilize their data to solve problems and derive more-informed conclusions. They rely solely on the human experience: The importance of individual experience cannot be overlooked but they don’t work best all the time, and minor details could be forgotten during the decision-making process. Now, experts use their best efforts to get accurate data on behalf of data scientists. Medianova CDN has millions of historical and real-time data and as Medianova R&D team we are implementing and presenting data-driven solutions.

What Are The Benefits Of AI?

There are also historically many techniques for time-series analyses including but not limited to AR, MA, ARIMA, SARIMA. However; such techniques are not trainable-algorithms and they are also timeconsuming. Today, we can see new approaches that focus on faster, more reliable, and scalable solutions using Machine Learning algorithms. In Medianova CDN; we use machine learning, and take actions more swiftly and intelligently.

Predicting Bandwidth Usage

As Medianova CDN we are getting millions of historical and real-time data from sensors located in servers. In an R&D effort; we constructed an Artificial Neural Network that predicts Bandwidth usage: Here, you can see a sample of our bandwidth usage data from one of our servers within a one-hour interval.

The Bandwidth Usage with one-hour interval.

Here’s the model with which we constructed an ANN that forecasts Bandwidth Usage.

Artificial Neural Network Model

The Used Artificial Neural Network Model including an input layer, four hidden layers, an output layer, and 1196 trainable neurons.

The following short-clip shows our ANN traffic prediction compared with actual traffic.

The Predicted Bandwidth Usage Video


The global CDN solutions in OTT, Gaming, E-commerce, Enterprise industries are being provided by Medianova CDN with the latest technologies. As Medianova R&D Team, we closely follow the latest technology trends and develop new technologies including intelligent solutions. Our customers do feel the difference. We are going to continue excellent solutions using machine learning and deep learning technologies. As said that Data is future and Medianova R&D Team is here.

Generic selectors
Exact matches only
Search in title
Search in content
Post Type Selectors