Predicting hypoxic hypoxia using machine learning and wearable sensors

Abstract The capability of detecting symptoms of hypoxia (i.e., reduced oxygen) and other cognitive impairments in-flight with wearable sensors and machine learning based algorithms will benefit the aviation community by saving lives and preventing mishaps. In this study, knowledge discovery processes were implemented to build classification models to predict hypoxia from wearable, dry-EEG sensor data collected from 85 participants in a two-phase study. Over a 35-minute period and while wearing aviation flight masks which regulated their oxygen intake, participants would alternate between a 2-minute cognitive test on CogScreen Hypoxia Edition and a 3-minute simulated flying task on X-Plane 11, with the oxygen concentration reducing every 5 min following the simulated flight task. The decrease in oxygen each 5 min simulated an increase in altitude. Features extracted from the EEG waveforms were transformed using principal component analysis to reduce the dimensionality of the data. Naive Bayes, decision tree, random forest, and neural network algorithms were utilized to classify the transformed brain wave data as either normal or hypoxic. The algorithms sensitivity ranged from 0.83 to 1.00 while the specificity ranged from 0.91 to 1.00. This study makes a step forward in developing a real-time, in-flight hypoxia detection system.