Russian scientists have developed a system capable of identifying deterioration in the quality of an internet connection in advance — even before it becomes noticeable to the user. The technology, created at Perm National Research Polytechnic University (PNRPU), provides forecast accuracy of up to 92.7% and allows operators to abandon costly subscriber surveys.
How the system "senses" internet problems in advance
The technology analyzes key network parameters in real time: ping, packet loss, as well as the volume of incoming and outgoing traffic.
Based on this data, the neural network determines whether the user will be satisfied with the quality of the connection or will experience discomfort.
Where and how it was trained
The AI model was trained in a complex environment where equipment simultaneously worked with wired, cellular, and satellite communications, constantly switching between channels.
Data was collected over five days. This allowed us to take into account different scenarios: peak loads, nighttime declines, and normal work activity.
What was discovered
During testing, scientists identified specific patterns.
Packet loss of up to 0.3% is almost imperceptible to the user. But after 1.8%, the quality of the connection deteriorates sharply.
It also turned out that the perception of the connection is strongly influenced not only by the average delay, but also by its stability. If the response time is constantly jumping, the user remains dissatisfied even with a high internet speed.
The university notes that such a system will allow operators to eliminate problems in advance — even before complaints appear. At the same time, the technology can be integrated into the infrastructure of providers and does not require the installation of additional expensive equipment, the scientists said.