Numerous studies have demonstrated that air pollution poses significant risks to human health, biodiversity, and the environment. In Germany, particulate matter has remained a persistent pollutant of concern since the 1990s, despite considerable improvements in air quality over the past decades. Consequently, air quality monitoring is required by law and remains a key tool for protecting public health. The resulting monitoring data provide a foundation for the development of deep learning models capable of predicting particulate matter concentrations and identifying their interdependencies with meteorological parameters. Hence, three neural network architectures are employed: feedforward neural networks (FNN), recurrent neural networks (RNN), and long short-term memory (LSTM) networks, each offering different capabilities for capturing temporal dependencies in air quality data. Meteorological parameters and temporal variables serve as model inputs, while the predicted particulate matter concentrations (PM10 and PM2.5) represent the model outputs. The analysis is based on hourly air quality and meteorological data from the LfULG and UBA monitoring networks, covering Dresden, Leipzig and Hamburg, three cities with markedly distinct climate, geographical and demographic profiles, over the period from May 2023 to January 2026.