The global prevalence of infectious diseases has emerged as a significant challenge in recent years.Surface transmission is a potential transmission route of most gastrointestinal and respiratory infectious diseases,w...The global prevalence of infectious diseases has emerged as a significant challenge in recent years.Surface transmission is a potential transmission route of most gastrointestinal and respiratory infectious diseases,which is related to surface touch behaviors.Manual observation,the traditional method of surface touching data collection,is characterized by limited accuracy and high labor costs.In this work,we proposed a methodology based on machine learning technologies aimed at obtaining high-accuracy and low-labor-cost surface touch behavioral data by means of sensor-based contact data.The touch sensing device,primarily utilizing a film pressure sensor and Arduino board,is designed to automatically detect and collect surface contact data,encompassing pressure,duration and position.To make certain the surface touch behavior and to describe the behavioral data more accurately,six classification algorithms(e.g.Support Vector Machine and Random Forest)have been trained and tested on an experimentally available dataset containing more than 500 surface contacts.The classification results reported the accuracy of above 85%for all the six classifiers and indicated that Random Forest performed best in identifying surface touch behaviors,with 91.8%accuracy,91.9%precision and 0.98 AUC.The study conclusively demonstrated the feasibility of identifying surface touch behaviors through film pressure sensor-based data,offering robust support for the calculation of viral load and exposure risk associated with surface transmission.展开更多
基金the National Natural Science Foundation of China(grant No.52108067).
文摘The global prevalence of infectious diseases has emerged as a significant challenge in recent years.Surface transmission is a potential transmission route of most gastrointestinal and respiratory infectious diseases,which is related to surface touch behaviors.Manual observation,the traditional method of surface touching data collection,is characterized by limited accuracy and high labor costs.In this work,we proposed a methodology based on machine learning technologies aimed at obtaining high-accuracy and low-labor-cost surface touch behavioral data by means of sensor-based contact data.The touch sensing device,primarily utilizing a film pressure sensor and Arduino board,is designed to automatically detect and collect surface contact data,encompassing pressure,duration and position.To make certain the surface touch behavior and to describe the behavioral data more accurately,six classification algorithms(e.g.Support Vector Machine and Random Forest)have been trained and tested on an experimentally available dataset containing more than 500 surface contacts.The classification results reported the accuracy of above 85%for all the six classifiers and indicated that Random Forest performed best in identifying surface touch behaviors,with 91.8%accuracy,91.9%precision and 0.98 AUC.The study conclusively demonstrated the feasibility of identifying surface touch behaviors through film pressure sensor-based data,offering robust support for the calculation of viral load and exposure risk associated with surface transmission.