Carbon-Aware Machine Learning: A Case Study on Cellular Traffic Forecasting with Spiking Neural Networks

Cellular traffic forecasting is an essential task that enables network operators to perform resource allocation and anomaly mitigation in fast-paced modern environments. However, the increasing amount of data collected by respective base stations makes their processing and analysis challenging. Machine learning (ML) algorithms have emerged as a powerful tool that can handle the large volumes of data and provide operators with accurate predictions. The environmental impact of such algorithms is, however, often overlooked in favor of predictive performance. Thus, a sustainable solution that takes into consideration the power efficiency of the considered ML models is crucial to mitigate the risks of the ongoing climate crisis shaping our future world. In this work, we utilize bioinspired spiking neural networks (SNNs) for the task of cellular traffic forecasting, as an answer to the power consumption challenge. We compare the spiking-based models against baseline architectures, i.e., Convolutional Neural Network (CNN) and Multi-Layer Perceptron (MLP), in terms of both predictive accuracy and associated energy demands using data collected from three different locations in Barcelona, Spain. Our results show that SNNs can lead to a decrease in energy costs while maintaining the quality of predictions.