The determination of possible anxiety by machine learning method from sound signals healthcare staff members who work at covid-19 clinics

Qualitative methods are generally used when evaluating individuals in the field of mental health. Although these questions are detailed and long, they can sometimes lead to false guidance because they proceed in a qualitative question-answer form. Health workers directly exposed to the COVID-19 disease during the pandemic have also started to feel threatened. Anxiety is a natural response of the body when one feels anxious and threatened. Anxiety affects emotions and thoughts as well as physical state. This study was conducted to assist healthcare professionals in the process of identifying anxiety by using output obtained from voice data. The State-Trait Anxiety Inventory (STAI) was used to measure anxiety levels. The participants were randomly selected from 30 healthcare workers actively serving in COVID-19 pandemic clinics and 30 healthcare workers not actively serving in COVID-19 pandemic clinics, all located in Ordu province. five verbal questions were asked via Phonic.ai and answers were obtained through voice responses, followed by answers to the STAI. SPSS package programs were used for the analysis of STAI data, and the results showed that the state anxiety levels of healthcare workers working in the emergency department were significantly higher. To create machine learning algorithms from voice responses, Python programming language was used The outputs of the State-Trait Anxiety Inventory (STAI) were compared with the outputs of the machine learning method. When the models were compared (VGG-19, ResNet50, AlexNet), the most successful one was found to be the 1D (dimensional) CNN (convolutional neural network) based model. In the emotion analysis of the model, an accuracy of 63.7% (63% sensitivity, 91% specificity) was achieved, while in the anxiety analysis, an accuracy of 73.5% (73% sensitivity, 94% specificity) was found. Based on these results, it is believed that the model developed will be more successful when supported by methods such as image and video recording in future studies.