G-Induced Loss of Consciousness Prediction Using a Support Vector Machine.

Area of Science:

• Aerospace Medicine
• Machine Learning Applications
• Physiological Monitoring

Background:

• Gravity-induced loss of consciousness (G-LOC) poses a significant risk to fighter pilots, potentially leading to fatal accidents.
• The brain can tolerate transient ischemia for 5-6 seconds under high +Gz exposure, known as the functional buffer period, without losing consciousness.

Purpose of the Study:

• To develop a machine learning model for predicting G-LOC within the functional buffer period.
• To evaluate the effectiveness of Support Vector Machines (SVMs) for G-LOC prediction.

Main Methods:

• 124 flight course students participated in the study.
• Linear soft-margin SVM, Gaussian kernel SVM (GSVM), and polynomial kernel SVMs were employed.
• Ten classifiers were developed at 0.5-second intervals (0.5-5.0s) post +Gz onset to predict G-LOC, using variables like age, height, weight, anti-G suit use, +Gz level, and cerebral oxygenation levels.

Main Results:

• The Gaussian kernel SVM (GSVM) demonstrated superior performance compared to other SVM models.
• GSVM achieved prediction accuracies ranging from 54.8% to 65.3% for classifiers from 0.5s to 5.0s.
• Specifically, prediction accuracy reached approximately 65% from 2.5 seconds after the onset of high +Gz exposure.

Conclusions:

• A machine learning approach using GSVM can predict G-LOC with notable accuracy within the functional buffer period.
• Further analysis with larger datasets and additional factors is recommended to enhance predictive accuracy for practical application in centrifuge training and flight.
• The findings suggest potential for improved pilot safety through early G-LOC detection systems.