Phone Videos Spot Work Impairment Early

Presenteeism—showing up to work while sick or mentally strained—drains productivity worldwide. Traditional surveys that ask workers how they feel take a lot of time and need experts to interpret the answers, which makes large‑scale monitoring hard. A new approach tries to solve this by turning everyday phone recordings into a quick health check.

Thirty‑eight employees filmed short clips on their phones for 10 to 30 seconds. Three clinicians watched each clip and gave it a three‑step rating: healthy, moderate issues, or unwell. The group consensus became the training label for an AI model that looks at voice, facial expressions, and what is said. The system uses a three‑stream attention framework that fuses these cues and predicts an ordered impairment score.

To see how well the model works, researchers ran a strict cross‑validation across many people. The best version, which mixes all signals, hit an overall F1 of about 0.77, an accuracy of 84 %, and an AUROC of 0.94, with a tiny calibration error. When the team compared models that included text versus those that did not, the text‑free versions fell far behind, proving that words matter. In a tricky group of brief talks from workers not labeled healthy, adding audio helped the prediction over text alone.

The findings suggest that a quick phone video can flag work‑function problems without a lengthy questionnaire. This could let companies spot early signs of presenteeism and intervene before productivity drops. The method is scalable, cheap, and works in real‑world settings.