ADEPT++: Extending Edge-Centric AAL Systems with Learning-Based Imputation and Synthetic Data Generation

This dataset contains multivariate time series recorded in 87 bedrooms under realistic Ambient Assisted Living conditions. It includes 608 sleep monitoring sessions (one per night), with each session representing uninterrupted data collection during the nighttime period from 20:00 to 12:00 the next day. Sensor readings were taken every minute, capturing the full duration of typical sleep cycles.

Each session records the following five environmental variables:

• Luminosity (lux)

• Temperature (°C)

• Humidity (%)

• Atmospheric Pressure (hPa)

• Noise Level (dB)

The recordings were acquired using a Samsung Galaxy S4 VE (GT-I9515), which integrates all sensors in a single device.

The total number of samples exceeds 317,000, spanning the entire calendar year with seasonal variability. The data exhibits significant heterogeneity across sessions in terms of duration, environmental conditions, and noise patterns, making it suitable for evaluating imputation models, anomaly detection, and time series analysis in AAL contexts.