Forum Summary

Many predictive algorithms are based on population averages and seek to represent the “average” person.  Some of these models, however, become much more useful when they are customized to fit individuals.  Predictive models generally fall into two categories: those based on first principles and those based on data.  Models based on first principles are usually easy to understand, but they only work if one possesses exact knowledge of the phenomena underlying the process being modeled.  Data-driven models require lessknowledge but can only be predict phenomena that they have already seen.  Researchers led by Jaques Reifman are currently using predictive models from both categories to model a variety of physiological processes. 

One of their projects involves modeling cognitive impairment caused by sleep deprivation.  They were able to customize a physiology-based model for specific individuals, allowing them to differentiate between people vulnerable to sleep deprivation and those who are relatively resistant.  Reifman’s group is also exploring a data-driven autoregressive model to predict blood glucose levels in diabetics.  Their goal is to create a model that will allow diabetics to maintain better control over their blood sugar levels.  Finally, Reifman and his colleagues are seeking to model the core temperature of members of the military in an effort to reduce the incidence of heat-related health problems.  All three projects involve predicting dangerous biometric changes before they take place.    

Post a Comment