Sustainable AI-based prediction of air pollution levels in London

Air pollution exposure not only leads to respiratory and cardiovascular diseases, but is also detrimental to cognitive abilities, mental health, and prenatal development. Thus, cities worldwide have invested in sophisticated air pollution monitoring systems to assess and reduce air pollution and its consequences. When excessive build-up of air contaminants occurs, emergency measures must be enacted to reduce human exposure and decrease pollution levels. Predicting such situations a few hours in advance is critical to prevent human health from being compromised. While usage of deep neural networks has become very popular, standard machine learning approaches remain very attractive: they deliver competitive performance, they do not rely on specialised equipment, and their energy consumption is sustainable. Experiments conducted on London air quality data demonstrate that Linear Regression achieves state-of-the-art performance, with 1-hour and 24-hour predictions displaying, respectively, 0.2 and 3.2 mean absolute errors. Moreover, its power usage is a fraction of what is required by its deep learning competitor for both training and predicting, i.e., 1/2840th and 1/126th, respectively. This is significant as they demonstrate air pollution prediction can be sustainable and accurate without prohibitive hardware investments.