The development of smart cities has been the epicentre of many researchers’ efforts during the past decade. One of the key requirements for smart city networks is mobility and this is the reason stable, reliable and high-quality wireless communications are needed in order to connect people and devices. Most research efforts so far, have used different kinds of wireless and sensor networks, making interoperability rather difficult to accomplish in smart cities. One common solution proposed in the recent literature is the use of software defined networks (SDNs), in order to enhance interoperability among the various heterogeneous wireless networks. In addition, SDNs can take advantage of the data retrieved from available sensors and use them as part of the intelligent decision making process contacted during the resource allocation procedure. In this paper, we propose an architecture combining heterogeneous wireless networks with social networks using SDNs. Specifically, we exploit the information retrieved from location based social networks regarding users’ locations and we attempt to predict areas that will be crowded by using specially-designed machine learning techniques. By recognizing possible crowded areas, we can provide mobile operators with recommendations about areas requiring datacell activation or deactivation.

The evolution of mobile communications, during the last decades, has led to a rapid increase in the number of users that mobile operators have to serve. To cope with this increase, mobile operators increment the number of base stations they are using resulting in an escalation of the corresponding energy footprint. This is why; the reduction of the total energy that is consumed from base stations has been the epicentre of many researchers. To achieve that, a common approach is to minimize the number of base stations that are used by activating only the necessary base stations without affecting the corresponding quality of service. In this paper, we present a method for predicting crowded areas based on machine learning techniques. The dataset used contains information about the number of users that have been connected to twenty base stations during the time period of 8 days. Prediction results can be used in order to make appropriate suggestions to mobile operators about bases stations that can be activated or deactivated. We propose a Probabilistic Neural Network and confirm its superior performance against two other types of neural networks.