Wearable sensors detect signs of escalation in autistic pupils minutes before disruption

A US study found biometric data could identify physiological changes two to five minutes before behavioural escalation became visible, though researchers caution that live classroom alerts have yet to be tested.

Researchers in the United States have found that wearable biometric sensors may be able to flag the approach of behavioural escalation in some autistic pupils minutes before disruption becomes visible, a result that could help schools intervene earlier and more calmly. In a study published by Frontiers in Psychiatry, the team monitored 30 students aged six to 14 across four rural public elementary schools and used machine-learning models to look for physiological patterns that preceded recorded incidents.

The research focused on signals that reflect the body’s stress response: heart rate variability, galvanic skin response, skin temperature and movement captured by a tri-axial accelerometer. Students wore a finger-based biometric ring during the school day over four weeks, while trained observers logged agitation, repetitive behaviour, vocal escalation, refusal, aggression, self-injury and attempts to leave the classroom. The study was designed to examine whether these measures changed before behaviour became obvious, rather than after the fact, and whether they could be turned into a practical warning system for teachers.

Among the algorithms tested, a long short-term memory model performed best, reaching 85.2% accuracy in distinguishing pre-escalation from non-escalation periods. The study also found that changes in physiology were detectable about two to five minutes before observable behavioural signs appeared, with the strongest predictors coming from heart rate variability and galvanic skin response. Adding movement data improved performance further, while skin temperature contributed more modestly. The authors said the pattern supports the idea that distress shows up in the body before it is visible in behaviour.

The findings are encouraging, but the paper is careful to frame them as retrospective rather than proof of a live alert system. The model was tested on previously collected data, not in real time, so it does not yet show that teachers could use such alerts effectively in a busy classroom or that they would reduce escalation in practice. Even so, the work adds to a growing body of research suggesting that wearable sensors could eventually help schools move away from reactive discipline and towards earlier, more individualised support for neurodiverse pupils.

The study also sits within a wider debate about how to support autistic children without relying too heavily on punitive responses or delayed observation. Its authors argue that physiological monitoring could complement behavioural frameworks such as Applied Behaviour Analysis by making internal stress more visible, while also aligning with cognitive behavioural approaches that focus on emotional regulation. They stress that any future classroom deployment would need strong privacy safeguards, individualised thresholds and clear staff training to avoid over-surveillance, false alarms or misuse of sensitive data.