Non-invasive sensor data from 50 households (motion, doors/windows and temperature)

The dataset originates from a non-invasive monitoring system deployed in 50 real-world households. This system comprises motion, magnetic, and temperature sensors designed to record the daily activity patterns of adults. The collected data were used to detect behavioural anomalies through convolutional autoencoder models, with the aim of enabling the early identification of behavioural changes, particularly in individuals with mild cognitive impairment (MCI). Each file in the dataset has been anonymised and organised into separate folders according to the sensor type. CSV files are identified using the naming convention {location}_{user ID}. Additionally, the dataset includes models trained using reconstruction error thresholds defined by the 90th, 95th, and 98th percentiles, covering both individual sensor models and a multisensor model for each type of sensor deployed in the households. The data collection was funded by the MIRATAR Project (TED2021-132149B-C41), supported by MCIN/AEI/10.13039/501100011033 and the European Union through the NextGenerationEU/PRTR initiative.