Urban development and public health depend heavily on air quality forecasts. Centralised data collecting, however, present difficulties for typical machine learning models and raises security and privacy issues. By facilitating model training across decentralised data sources while maintaining data privacy, federated learning (FL) presents a possible option. Using information from many Internet of Things (IoT) sensors placed around cities, this research offers an FL framework for forecasting the Air Quality Index (AQI). By preserving data privacy at each source, the suggested FL framework makes it easier to train models using a variety of sensor data. To minimise the need for data transfer, only model updates are communicated centrally once initial data processing and model training at the sensor level are completed using local computing resources. Comparable accuracy to centralised methods was attained by the FL model, which also improved data privacy. This study offers a scalable, real-time, and privacy-aware framework for air quality monitoring systems that use the Internet of Things technology, which is a major improvement for smart city efforts and environmental monitoring.