Furno et al. describe a framework for automated demand profiling in mobile networks. Since participants in mobile communications can move around, both spatial as well as temporal characteristics of network traffic must be considered. In addition, users utilize the networks for voice communication, texting, and data traffic. In their framework, the authors extract snapshots of network traffic identified through time intervals, geographical areas, and traffic type. These snapshots are arranged in a graph, where snapshots are connected by weighted edges that represent similarity measures such as traffic volume similarity and traffic distribution similarity.

The snapshots are partitioned into similar mobile traffic conditions through a clustering algorithm. To test the framework, the authors used two datasets, the Orange 2013 D4D Challenge and the Telecom Italia 2014 Big Data Challenge. In both cases, they successfully identified typical behaviors, such as workday versus weekend and daytime versus nighttime. More challenging is the categorization of outliers representing unusual events such as network outages, power failures, holidays, or sporting events. While public holidays applied to all covered locations, there were spatial temporal or spatial variations in user behavior as well. This was more prominent, however, in spatially and temporally constrained events, such as high-profile soccer matches or cultural events. In both datasets, the framework successfully associated such events with categories other than what would be expected during the respective time and location, and with the same categories for events with similar user behavior.

With suitable infrastructure and allocation strategies, such approaches can help with the more efficient use of network resources. User behavior and expectations will probably also change, and sufficient network capacity to satisfy everybody’s demands at all times still may be beyond reach.