Support Vector Machines to improve physiologic hot flash measures:
Application to the ambulatory setting
People
Abstract
Most midlife women experience hot flashes. The conventional criterion for classifying hot flashes from physiologic signals (=2 µmho, 30 sec) has shown poor performance. In the laboratory, we improved this measure’s performance using Support Vector Machines (SVMs), an advanced pattern classification method. The primary aim was to compare the conventional criterion to SVMs to classify hot flashes from sternal skin conductance in the ambulatory setting. Thirty-one women with =4 hot flashes/day underwent 24 hours of ambulatory sternal skin conductance measurement. Hot flashes were quantified with conventional and SVM methods. Conventional methods had poor sensitivity (sensitivity=.57, specificity=.98, positive predictive value (PPV)=.91, negative predictive value (NPV)=.90, F1 score=.60) in classifying hot flashes, with poorest performance among women with high body mass index (BMI). SVM models showed improved performance (sensitivity=.87, specificity=.97, PPV=.90, NPV=.96, F1=.88) and reduced variation by BMI. SVMs can improve the performance of ambulatory skin conductance hot flash measures.
Citation
Copyright notice