Device-to-device (D2D) communications play a key role in today's communication infrastructures. They help optimise network capacity, reduce latency and enhance quality of service. The main difficulty in establishing clusters lies in the ability to optimally designate a CH (cluster head) for each group. In this study, we present a method for selecting cluster heads (CHs) and forming clusters for device-to-device exchanges. Unlike methods that focus primarily on geographical proximity or data similarity, our proposal incorporates both the geographical position of users (smartphones) and their residual energy, thus offering a content-free structure that is more appropriate for dense and varied networks. The strategy begins with a preliminary classification using the K-means algorithm, followed by an optimised CH selection process within each cluster. Observations from multiple simulations show that the suggested method provides satisfactory results with a substantial 39.15% increase in throughput. Furthermore, taking energy into account when selecting CHs extends the longevity of the devices and the stability of the network.