| Keywords |
Activities of Daily Living; Anomaly Detection; Sleep Pattern; Occupancy Detection; Non-intrusive Sensor |
| Abstract |
With population aging and the increase in single-person elderly households, continuous monitoring of daily living patterns has become increasingly important for supporting older adults and vulnerable populations.
This study proposes a non-intrusive Activities of Daily Living (ADL)-related anomaly detection framework using indoor environmental sensor data. The proposed framework estimates sleep patterns and room-stay patterns from temperature, humidity, CO , particulate matters (PM 2 2.5, PM10), VOCs, illuminance, and PIR (Passive Infrared) data collected by IoT sensors, and determines ADL-related anomalies by comparing daily patterns with individual baseline distributions. A sleep detection model was developed using 15-min sensor features, temporal contextual variables, and manually recorded sleep labels. The model was implemented as an ensemble classifier combining LightGBM and ExtraTrees, and achieved an F1-score of 0.93 in sleep/non-sleep classification. For behavior pattern analysis, room-level occupancy outputs were converted into daily behavior sequences consisting of room-stay tokens, and deviations from individual baseline routines were quantified using normalized edit distance. Sleep anomalies and behavior sequence anomalies were then integrated using a rule-based decision structure. The proposed framework was further applied to 100 single-person elderly households that was not included in the model training process. The proposed method was able to detect a case of a behavior pattern anomaly, which include no valid sleep episode within the sleep evaluation window, and missing behavior sequences in the selected household. The observed event was later confirmed as hospitalization of the resident. These findings suggest that non-intrusive indoor environmental sensors may support the early detection of ADL-related changes without requiring wearable devices or camera-based monitoring. |