Development of an Atopic Dermatitis Incidence Rate Prediction Model for South Korea Using Air Pollutants Big Data: Comparisons Between Regression and Artificial Neural Network

Abstract

We have developed models to predict the incidence of atopic dermatitis using regression analysis and artificial neural networks (ANN). Initially, the prediction models were created using various inputs, including air pollutants (SO₂, CO, O₃, NO₂, and PM₁₀), meteorological factors (temperature, humidity, wind speed, and precipitation), population rates, and clinical data from South Korea, referred to as the average model. Subsequently, we developed models that use sex and age as variables instead of population rates, named the sex and age model. Both sets of models were designed to forecast incidence rates on a nationwide scale (NW), as well as for 16 administrative districts (AD) in South Korea, which includes seven metropolitan areas and nine provinces. We found that SO₂ significantly affected the incidence rate, and the inclusion of regional variables in the AD models helped account for regional variations in incidence rates. The average models generally provided accurate predictions of incidence rates, with SO₂ chosen as the key independent variable in the regression models for the five air pollutants studied. The R² values for the average models using regression are 0.70 for the NW model and 0.89 for the AD model. Among the ANN-based models, the R² values are 0.84 for the NW model and 0.90 for the AD model, this indicated a slightly higher predictive accuracy. For the sex and age models, we differentiated between children under 10 years of age and those older. In these models, ANN demonstrated greater accuracy than regression, with R² values of 0.95, 0.92, 0.96, and 0.92 for the sex and age NW model under 10 years old, sex and age AD model under 10 years old, sex and age NW model over 10 years old, and sex and age AD model over 10 years old, respectively.