A Proposed Deep Learning Framework for Air Quality Forecasts, Combining Localized Particle Concentration Measurements and Meteorological Data

Air pollution in urban areas has significantly increased over the last few years due to industrialization and overpopulation. Therefore, accurate predictions are needed to minimize their impact. This paper presents a neural network-based approach for Air Quality Index (AQI) forecasting, applying two different models: a variable depth neural network (NN) called slideNN, inspired by the capabilities of a Recurrent Neural Network (RNN) and a Gated Recurrent Unit (GRU)-based model. Both models use past particulate matter measurements alongside local meteorological data as inputs. Then, this research explores the practical application of multi-depth and recurrent neural networks in air quality forecasting, providing a hands-on case study and model evaluation for the city of Ioannina, Greece, targeting the appliance of such models either on edge devices or as cloud services. SlideNN variable-depth architecture consists of multiple independent neural sub-model strands varying in size, enhancing feature extraction and predictive accuracy. At the same time, the GRU model is examined on its ability to capture temporal dependencies in the data. Finally, both models are combined to offer a cloud-based high-precision hybrid. Experimental results show that the GRU model consistently outperforms the variable-depth architecture regarding forecasting losses. In contrast, complex hybrid GRU-NN models outperform both, delivering additional localized information that can be exploited by established particle concentration map monitoring services.