Artificial intelligence analysis supported investigation of the effects of indoor air quality on health

Among the health effects associated with indoor air pollutants in schools, the increased risk of respiratory tract infections (RTI) and Sick Building Syndrome (SBS) have an important place. This study aimed to determine the effect of indoor air quality on the incidence of respiratory tract infections, sick building syndrome, and absenteeism in a high school, to design an air quality measurement system to estimate the air quality index of indoor spaces, to develop a deep learning network as an artificial intelligence-based model with data obtained from sensors. Two workshop classes were selected using the simple random method. During a total of 118 working days, carbon dioxide (CO2), temperature, humidity and particulate matter (PM) were measured with an air quality measurement system whose software and electronic design were specially prepared for this study. Following the completion of the measurements, a survey form was administered to the students. The first part of the survey included the sociodemographic data form and the questionnaire regarding RTI prepared by the researchers, and the second part included the MM 060 NA School scale to detect SBS symptoms. In addition to the survey information, absenteeism information of the sampled students was provided. Among a total of 79 students, the prevalence of SBS was 8.9%, and the number of people who stated that they had RTI at least once was 14 (17.7%). The annual average values across the school were measured as 47.6% for relative humidity, 24.8°C for temperature, 1433.7 ppm for CO2, 15.1 μg/m3 for PM1, 22.3 μg/m3 for PM2.5, and 31.1 μg/m3 for PM10. In the logistic regression analysis, it was determined that smoking increased the risk of having an RTI by 8.0 times. A low positive correlation was determined between illness-related absences and all indoor air quality variables except humidity and temperature. It was observed that absences due to illness were higher in the days following the days when CO2 and PM2.5 exceeded the limit. Regarding AI-based prediction models, lower error rates were achieved in LSTM, achieving an accuracy rate of 81.25% for both datasets. No significant effect of indoor air quality on the development of RTI and SBS was found, but it was observed that the increase in CO2 and PM increased sickness-related absences at a low level.