Handover prediction in 5G networks

Developing models for more accurate 5G handover prediction.

Challenge

Every mobile device in the network is served by a specific base station. When a user moves from one station to another, their device must switch to a new station at some point. This procedure is called a handover. Traditionally, the decision to start a handover is made on the basis of strict rules — for example, when the signal from the new station is stronger than the signal from the old station for a certain amount of time. However, these rules are not always optimal, and sometimes the handover leads to signal degradation, or conversely, the handover takes too long to initiate. The task of this project was to determine the perfect moment of handover based on machine learning algorithms.

Solution

Collection and preparation of data from mobile devices and service stations was carried out. On the basis of several dozen different parameters contained in the data, features were formed to determine the beginning of a handover. Several ML models were tried, such as linear models, gradient boosting, and recurrent and convolutional neural networks, out of which the best was chosen.

Results

A model was developed to predict the signal strength from the old and new station after a given amount of time. This made it possible to predict the appropriate handover moment more accurately, thereby improving the user experience of mobile network users.