Smart-Home Engineering for Social Housing
Overview
In partnership with a UK social landlord and the University of Glasgow, I led the engineering of two IoT systems studies on a real, occupied social-housing testbed. Both target affordable, privacy-preserving sensing for healthier living conditions and safer ageing in place.
System 1: Multi-Sensor Mould-Risk Forecasting
A real-time IoT system that forecasts mould risk 6–48 hours ahead in social housing, enabling tenants and housing providers to take preventative action (ventilation, heating adjustment, dehumidification) before conditions worsen.
Engineering results
- 24-hour AUC-ROC: 0.851 to 0.957 across two real homes.
- Cross-home transfer AUC: up to 0.968.
- Multi-sensor fusion: temperature, humidity, CO₂, ventilation proxies, heating signals.
Manuscript: “Multi-Sensor IoT Fusion for Mould Risk Forecasting in Social Housing” — under review at IEEE Internet of Things Journal.
System 2: Non-Intrusive Welfare Monitoring for Lone Elderly Residents
A privacy-preserving IoT system for monitoring the welfare of lone elderly residents — without cameras, microphones, or wearables.
Engineering results
- F1 scores: 0.848 and 0.814 across two homes.
- Detection rates: 87% to 100% across clinically motivated scenarios.
- Cross-home F1 transfer: up to 0.841.
- Sensing cost per flat: ~£200–300.
- Rapid deployment: usable from just 7 days of baseline data.
Sensing approach
- Per-appliance power monitoring.
- Environmental sensing (temperature, humidity, light).
- Radiator valve state inference.
- Circuit-level energy signals.
Manuscript: “Multi-Signal IoT Anomaly Detection for Non-Intrusive Welfare Monitoring of Lone Elderly Residents” — under review at IEEE Journal of Biomedical and Health Informatics.
Technologies Used:
- Python
- PyTorch
- ESP32
- LoRaWAN
- MQTT
- Time-Series Analysis