CDDM: A method to detect and handle Concept Drift in Dynamic Mobility Model for seamless 5G Services

In 5G network, it is required to provide multiple high bandwidth services to the consumers with minimum latency and to achieve this advanced provisioning of resources are required. In order to advanced provision resources, we need to predict the mobility patterns of the consumers so that this can be done in an efficient fashion. To predict the mobility pattern of the user equipment's (UE's) one tends to fit a single ML/AI model and deploy it in the network. In real life, there can be more than one underlying pattern in the data and a single model need not be necessarily generalizable. Also, in some cases, the underlying mobility pattern may change with time which is known as concept drift. If the distribution change is beyond a threshold, it is better to retrain the model to generalize it for the new set of data and it is a costly exercise. However, in the case of multiple models, the problem is complex, and confusion arises on which model to retrain when the concept drift is detected. In this work, we propose a new method to handle concept drift in the multiple model scenario and further advise which model to retrain once the concept drift is detected. This mechanism can be used to generate accurate results without costly retraining on the model multiple times even when the underlying distribution changes rapidly. Results on simulated data demonstrate the efficacy of the proposed method and are able to give accurate mobility predictions with minimum retrainings.