This paper is useful because it moves beyond the question of whether a wearable can detect change and asks a more design-relevant question:

When is it actually appropriate to intervene?

Rahman et al. use a Just-in-Time Adaptive Intervention (JITAI) framework to decide whether a behavioural prompt should be delivered based on more than one signal.

That is very close to the architecture I am now developing for the airport project.

The strongest takeaway for me is:

Do not trigger support from one body signal. Combine physiology with behaviour, environment and previous responses, and suppress prompts when the moment is inappropriate.

What the paper actually studied

The researchers developed a context-aware personalised nudging system aimed at reducing sedentary behaviour.

The system used a custom smartwatch application in a free-living observational study with 18 participants at the University of Delaware.

Rather than triggering a prompt simply because the wearer had been sitting for a certain amount of time, the model incorporated several types of information:

• temporal factors — when the event was occurring;
• physiological state;
• behavioural indicators — such as sitting or walking;
• environmental / contextual attributes — including location;
• historical information — including the outcome of previous interventions.

A decision-tree model then used these features to estimate whether a nudge was likely to produce the intended behavioural response.

Context improved the decision to prompt

The paper reports that adding contextual features improved classification of nudge outcomes from 0.42 to 0.78 accuracy across 787 sitting bouts.

A separate walking-nudge model achieved 0.70 accuracy across 207 walking bouts.

The authors also report that walking bouts following nudges were longer, covered greater distances and had higher average speeds than non-nudged bouts.

The important point for my project is not the sedentary-behaviour outcome itself.

It is that the system did not treat one sensor reading as enough information to decide what to do.

Design principle

A physiological signal should be one part of the decision about whether a prompt might be useful, not the sole trigger for an intervention.

This is close to the airport problem

An airport journey is full of situations where the same physiological change could mean very different things.

For example, an increase in pulse or EDA could happen while somebody is:

• walking quickly to a gate;
• carrying luggage;
• standing still in a crowded queue;
• going through security;
• waiting for delayed information;
• drinking coffee;
• or dealing with an unexpected gate change.

A system that reacts to the body signal alone cannot distinguish these situations.

This paper strengthens the argument that I need a wider decision layer:

physiology + movement + airport context + previous responses

rather than:

physiology → automatic prompt

The previous response matters too

One of the parts I find particularly useful is the inclusion of historical intervention outcomes.

The system considers not only what is happening now, but what happened after previous prompts.

That opens an important direction for my own project.

If the traveller repeatedly dismisses prompts in a particular situation, the system should probably become less likely to interrupt in similar situations.

Likewise, if a particular type of prompt or support is repeatedly selected in a particular context, that may become useful information for future personalisation.

This does not mean the system learns the person's internal state.

It means it can learn something much narrower and more defensible:

when this particular person tends to find a prompt useful or unhelpful.

Not every technically valid trigger should become an interruption

This paper makes the timing problem clearer.

A physiological change may be real, but a prompt may still be inappropriate.

For example, during:

• security screening;
• passport control;
• boarding;
• a conversation with airport staff;
• or another high-attention task,

a check-in could become another demand.

So I am increasingly thinking of the system as having two decisions rather than one:

1. Has something changed enough to justify attention?
2. Is this an appropriate moment to ask the traveller?

Important interaction distinction

Detection and interruption are separate decisions.

The system may notice a change but deliberately wait, suppress or delay the prompt if the current task makes interruption inappropriate.

This feels particularly important for autistic airport accessibility because the prompt itself could otherwise become part of the sensory or cognitive burden.

JITAI gives me a useful framework

The idea of a Just-in-Time Adaptive Intervention is useful to me even though my project is not a health-behaviour intervention.

The transferable design logic is:

observe → interpret context → estimate receptivity / appropriateness → prompt only when useful → learn from response

For my system, I would adapt that into:

personal baseline + physiological change + movement + airport context + previous user choices → cautious interpretation → check whether prompting is appropriate → transparent check-in → traveller decides

That is more sophisticated than the earlier version of my system where physiological change itself was the main trigger.

What it DOES establish

For my project, this paper provides useful evidence that:

• wearable sensing can be combined with contextual and historical information in a free-living system;
• adaptive prompts can be timed using more than one feature;
• adding context can improve the system's ability to predict whether a prompt will produce the intended response;
• previous intervention outcomes can contribute to future personalisation;
• and context-aware prompting is technically feasible outside the laboratory.

What it does NOT establish

This study does not show:

• that physiological signals detect sensory overload;
• that this nudge model transfers directly to autistic adults;
• that an airport support prompt will improve accessibility outcomes;
• that the same context features will matter in an airport;
• that an algorithm should decide what support the traveller needs;
• or that a successful behavioural nudge is equivalent to a useful accessibility intervention.

It is also a small study with 18 participants, so I should treat the model as a useful precedent for architecture and timing rather than proof of general effectiveness.

What this means for my airport system

This paper changes the system logic in an important way.

My earlier architecture was roughly:

physiology + movement + airport context → interpretation → prompt → user choice

I now think it needs another stage:

physiology + movement + airport context + previous responses → cautious interpretation → is this a good moment to prompt? → transparent check-in → user choice

That small addition matters.

It means the system is not only trying to avoid misreading the body.

It is also trying to avoid interrupting the person at the wrong time.

What changed in my thinking

This paper makes me think the next design problem is no longer simply how to detect change.

It is:

How should the system decide when a detected change is worth interrupting the traveller for?

That introduces a new layer of interaction design around:

• receptivity;
• task demand;
• prompt frequency;
• previous dismissals;
• journey stage;
• and manual control.

For me, this is a significant refinement because it makes the system feel less like a reactive sensor and more like a context-aware support layer.

The strongest takeaway is:

The best prompt may sometimes be no prompt at all.

Source

Rahman, T., Vemuri, A., Firkin, C.J., Bodt, B., Orsega-Smith, E., Dominick, G.M. and Decker, K. (2026). Improving Context-Aware Personalized Nudging: Using Wearable Sensors to Reduce Sedentary Behavior. Delaware Journal of Public Health, 12(1), pp. 106–112.

Key points from my reading: JITAI framework; 18-person free-living study; smartwatch-based motion, physiological and contextual sensing; decision-tree model; temporal, physiological, behavioural, environmental and historical features; nudge-outcome classification improving from 0.42 to 0.78 across 787 sitting bouts; walking-nudge accuracy of 0.70 across 207 bouts; direct relevance to context-aware prompt timing rather than single-signal triggering.