Back

Human-Perception-Aligned Machine Learning for Indoor-Outdoor Classification

Mahfoud, D.; Tang, Z.; Najjar, R. P.

2026-01-09 physiology
10.64898/2026.01.08.698168 bioRxiv
Show abstract

Daylight is vital for eye, brain, and overall health, yet misperceptions of what constitutes true outdoor daylight exposure can lead to poor behavioural choices and weaken public health recommendations. We developed InNOut, a perception-aligned machine learning model trained on 73,879 minutes of multispectral light data to classify environments as indoor or outdoor, and benchmarked against public (n = 383) and expert (n = 17) judgments across 183 scenes. InNOut achieved excellent performance (AUC = 0.92 [95% CI, 0.92-0.93], sensitivity 73.9%, specificity 94.5%), closely aligning with expert (83.5%) and public (80.8%) judgments. Clear outdoor and indoor scenes were reliably classified, whereas ambiguous settings (e.g., windowed rooms, vehicles) were often judged indoor by participants but outdoor by the model, consistent with their spectral light profiles. InNOut bridges perception and measurement, offering a scalable means to map light environments and inform digital interventions in medicine and public health.

Matching journals

The top 6 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.