Short-Lead Routine Discordance Detection in PIR-Based Smart Homes Using Transformer Occupancy Prediction

Passive infrared (PIR) motion sensors are frequently used in AmbientAssisted Living (AAL) because they are low-cost, unobtrusive, and less privacy- invasive than camera- or microphone-based monitoring. However, PIR sensors provide sparse binary activations rather than direct activity labels or continuous occupancy information. This study proposes a privacy-preserving home- monitoring pipeline that converts PIR events into room-level occupancy states using a Finite State Machine (FSM) and then applies a transformer encoder to predict multi-room occupancy at a 5-min lead time. The median sampling interval was 0.483 min, corresponding to approximately ten samples for the 5-min predic-tion horizon. The model achieved a training binary accuracy of 89.5% and a validation binary accuracy of 87.2%, with final training and validation losses of 0.273 and 0.325, respectively. On the held-out test segment, the transformer achieved higher binary accuracy and lower binary mean discordance than a persistence baseline, while further room-level and imbalance-aware evaluation is left for future work. Habitual discordance was computed as the mean absolute deviation between predicted and observed 5-min-ahead room-occupancy vectors. The held-out 24-h test window showed how short-lead discordance can be visual-ized as an interpretable candidate flag for routine deviation. These results support the feasibility of transformer-based, short-latency discrepancy monitoring in PIR-based AAL settings.