Non-skin-contact wearable + IBWO-tuned Random Forest for personalised thermal comfort prediction

Design Solution · HVAC & Energy

Design Solution · Dream about it

ML-tuned wearable microclimate sensing + Random Forest classifier for individualised ASHRAE thermal comfort prediction in offices.

A non-contact wearable device captures ambient microclimate data (temperature, humidity, air velocity, radiation) and feeds it into a Random Forest model whose hyperparameters are optimised via Improved Beluga Whale Optimisation (IBWO) to predict individual thermal sensation votes in real time. The approach addresses the gap between fixed-setpoint zone HVAC control and inter-personal thermal heterogeneity in shared office spaces, enabling adaptive, occupant-centric comfort targeting without skin contact or survey burden.

This solution pairs a non-contact wearable device measuring ambient microclimate (temperature, humidity, air velocity, radiation) with a Random Forest classifier whose hyperparameters are tuned via Improved Beluga Whale Optimisation, aiming to predict individual thermal sensation votes in real time and enable HVAC systems to target occupant-centric comfort across a shared office zone. The problem it addresses — fixed-setpoint zone control ignoring genuine inter-occupant thermal heterogeneity — is well established in building performance research and productivity-linked workplace literature. The lab validation reports over 98% accuracy on a small cohort, but no field trial, independent replication, or live building deployment exists; generalisation from a lab cohort to different building typologies, climates, and occupant demographics without retraining is unvalidated. Metabolic rate, clothing insulation, and activity level — key ASHRAE comfort variables — are not captured by microclimate sensing alone, reducing the model's contextual completeness. There is also no demonstrated integration pathway with existing BMS or HVAC control systems, and the privacy implications of continuous wearable biometric sensing in the workplace under GDPR have not been addressed. The potential value, if real-world accuracy holds, is meaningful for high-occupancy workplace environments where thermal complaint is a known productivity and retention issue; the honest position is that the lab result is interesting but the gap to deployable, consented, BMS-integrated product in a live building is large and largely uncharted.

Strengths

• Predicts individual thermal comfort without occupant surveys or feedback loops, removing behavioural burden.
• Bridges fixed-zone HVAC control with inter-occupant thermal variation, enabling granular comfort responsiveness.
• Non-contact design eliminates skin barrier concerns and improves wearability acceptance in office settings.
• IBWO hyperparameter tuning potentially optimises classifier performance beyond standard Random Forest tuning methods.

Considerations

• Requires continuous wearable device carry and charging; adds personal tech overhead and user compliance burden. (moderate)
• Microclimate sensing alone cannot capture individual metabolic rate, clothing insulation, or activity level—key ASHRAE inputs—reducing model contextual completeness. (high)
• Random Forest is opaque; difficult to audit which microclimate factors drive comfort predictions, limiting trust and diagnostic utility for HVAC operators. (moderate)

Risks

• Laboratory validation at >98% accuracy on small cohort; no published field trial, independent replication, or live building deployment evidence—generalisation and real-world performance unknown. (high)
• No evidence of integration pathway with existing BMS/HVAC control systems; practical deployment, latency, and system interoperability unvalidated. (high)
• Model trained on lab cohort may not transfer to different building typologies, climates, or occupant demographics without retraining—potential overfitting risk. (moderate)
• Privacy implications of continuous wearable biometric sensing in office environment not addressed; potential regulatory/consent risk under GDPR or workplace privacy law. (moderate)
• IBWO optimisation method is emerging; no comparison against standard Bayesian or grid-search tuning or published ablation of its added value—algorithmic novelty not empirically justified. (low)

Performance

• Claimed prediction accuracy: >98% (lab conditions, small cohort)
• Sensor modality: Ambient microclimate only — no skin contact required
• ML optimiser: Improved Beluga Whale Optimisation (IBWO) + Random Forest
• Deployment evidence: None — controlled experiment only

Reality check

Laboratory accuracy claim (>98%) is derived from controlled, single-cohort trials only. No peer-reviewed field validation, cross-site replication, or deployment in live buildings found. IBWO algorithm is a metaheuristic optimization routine published in academic literature (not a novel method). Claim of feeding predictions 'back to zone HVAC setpoints' is conceptual; no evidence of actual control loop integration or energy impact measurement. Wearable form factor and non-contact sensing reduce friction vs. skin-contact thermal sensors, but this alone does not validate real-world accuracy or ROI.

#machine_learning #thermal_comfort #wearable_sensing #occupant_wellbeing #adaptive_hvac #ASHRAE #random_forest #personalisation

Source