Hi Jacky, your idea is an insightful & interesting one. iButtons are great for both resolution and precision, but I don't know of a way to continuously stream the data, or to be able to attach it to an alarm. I'm still looking to see what device(s) seem to be the best bet, but here's what I know so far, and I will update this thread as I find out more. Let me know if you have any more questions.
-A device that will send an alarm to your (phone/computer) when temperature it sense rises above a user-specified set point seems best. This would be potentially useful if you can first do some beta testing with ibuttons to see if you can characterize the temperature rise your son experiences with an episode. It probably won't be exactly like the one shown in the paper, as body temperature fluctuations can differ significantly between people. Most research articles will show you the averages- not the individual profiles. I am still looking for an alarm system with < .5 degree accuracy (most seem to only do about a degree).
-As you know, the most important thing when picking out iButtons or an alarm system would be the sensitivity and accuracy of the temperature sensor (as the entire range of body temperature fluctuation is only a couple of degrees, precision to the 10th or more of a degree is important). It will be more expensive, but it's necessary to detect the wake to sleep transition.
Based on the paper you found, and the other ~10 papers on the topic I am seeing, both rising temperature and rate of temperature change might be useful places to start looking for precise trends. Do you have any temperature data yet that you have looked at? I'm not an expert data scientist, but I do have some experience with temperature and sleep data and would be happy to take a look if you are trying to identify features you might be able to use to predict an episode.
As OP_Eng mentioned, devices like the smart cap claim to be able to detect when the user is nodding off with small EEG sensors. I don't know how accurate it is but it may be worth a try. The other truck driver devices seem to measure things like head tilt (with mixed reviews). It might be a simpler place to start.