Cellular-based connections are growing exponentially. The proliferation and worldwide shipment of IoT (Internet of Things) devices is continually increasing. By January 2022, the customer reported that the percentage of IoT devices as part of the total inbound roaming had grown to no less than 70%. The detection of these devices was primarily based on declared IMSI ranges or APN lists from discount agreements or based on the hardware profile of the device, the IMEI information.
However, for outbound roaming, although the customer had assigned some specific IMSI ranges for IoT, the legacy IoT devices were not appropriately organized and were difficult to map. In addition, the customer could not rely entirely on the IMEI information present on received TAP files from the roaming partner as this was an optional field, not always populated.
The challenge was to gain complete visibility of the IoT traffic, inbound and outbound, to detect potential business opportunities.
The customer agreed on a Proof of Concept (PoC) to test TOMIA’s IoT detection based on Machine Learning. The AI/ML algorithm is mainly used when IMEI information is not available. It can interpret signaling and usage information to provide insight into human and IoT traffic patterns, including silent IoT devices.
The exercise consisted of removing the IMEI information from a large amount of traffic, running the AI/ML algorithm in a controlled manner, and comparing the results. The activity was supervised by a data scientist who perfected and trained the models. The main goal was to increase the detection accuracy on the files that the IMEI was missing based on the known IoT traffic patterns.
The classification model reached an accuracy of over 95%, which was accepted by the customer. IoT detection based on AI/ML can support operators in finding hidden devices and building more accurate IoT business cases. It can also be used as a second validation layer to monitor the accuracy of IoT discount agreements considering their IoT customer segmentation and traffic split.