Dengue Fever Screening Using Vital Signs by Contactless Microwave Radar and Machine Learning

In order to carry out high-precision and low-cost infection screening, our group is working on the development of a contactless infection screening system using multiple vital signs detected from a microwave sensor. In this study, we focused on sensor signal processing to extract respiration and heart beat signals with high accuracy and constructed a machine learning model suitable for screening infectious diseases using Neural Network and SoftMax function. The effect of respiration while measuring the heart rate was reduced by correcting the phase shift of the respiration signal and subtracting the respiratory component measured from the microwave sensor. The machine learning model of Neural Network was Logistic Regression using SoftMax function as an activation function. A total of 158 subjects (79 normal subjects, 79 infected subjects) were tested with the proposed system. Since the SoftMax function can provide a stochastic expression as output, it is now possible to not only determine whether the disease is an infectious one but also to express the probability of infection. We obtained a highly generalized training model by evaluating the generalization of the training model with the cross-entropy loss function and achieved a classification accuracy of 98%.