Introduction to Predictive Biometric Systems - Part 5: What does a successful predictive biometric pilot look like (a case study with heat exhaustion)?

February 10th, 2020

The last post went relatively in-depth into the difference between set-value predictive biometric systems and self-learning predictive biometric systems

It’s always much easier to understand something with an example, so now we’re going to use a case study from Canaria Technologies for heat exhaustion prediction. In this case study, a client conducted a heat exhaustion pilot for 6 weeks: this case study covers the first part of a two-part pilot. It explains the process of installing and testing a predictive biometric system using the Canaria Puck on a workforce who has expressed recurring heat exhaustion problems in a site in Northern Australia.

How can Canaria Technologies Detect when a Heat Exhaustion Incident Has Taken Place?

Of course, before you can predict something, you have to be able to detect it first. So how do we go about detecting a heat exhaustion event in one of our users?

The pedantic definition is that we’ll have captured the symptoms of a heat stress incident (it’s not always possible to cross-reference heat stress incidents with self-reporting data that an incident has taken place: good data science practice never claims to have captured an event until a user has self-reported an incident as well as the symptoms having been recorded). These include:

  • Gait analysis (someone fainting from heat exhaustion in an extreme situation or struggling to walk normally from dizziness)
  • Abnormally high skin temperature compared to ambient temperature (ie. 40C compared to 38C ambient temperature)
  • Raw PPG readings showing abnormal physiological stress
  • Users going into a cool environment and recovering by being still after a sharp increase in their skin temperature and physiological stress readings
  • Elevated respiration rate
  • Elevated heart rate heatexhaustionVheatstroke


A set-value predictive biometric system for heat exhaustion is the first step to deploying a successful predictive biometric system that be trusted to give reliable alerts for users. Puck_from_above

For the first step, alarms are active in the field but set to silent to allow for in-depth threshold testing for the first group of users. These thresholds are based on a ‘best educated guess’ drawing from a combination of medical studies published in scientific papers and the pre-existing data sets created by Canaria Technologies previously. The ‘switch on’ point for the alarms is when the system hits approximately 85% accuracy. This avoids the common pitfall of early users for new alarm technologies getting ‘alarm fatigue’ from too many false positive alarms during the pilot and early deployment phases.