This study addresses the problem of air quality in homes, focusing on the analysis and research of environmental patterns linked to different domestic activities. Based on a real-world dataset composed of sensor data collected in real time from six homes over a 14-day period, this research aims to identify how daily activities impact indoor air quality using two different approaches: the first in which each household uses its own data to train its own models. The second scenario in which there is a federated collaboration without sharing the data. In particular, through the definition of a detailed methodology, a pipeline is described that starts from data preprocessing, followed by the application of machine and federated learning techniques including the use of K-Means with Dynamic Time Warping (DWT) to detect and associate patterns of air quality variation, leading up to the validation of the results.