Comparing classic regression with credit scorecard model for predicting sick building syndrome risk: A machine learning perspective in environmental assessment

Abstract

Quarantine policy for COVID-19 control brings outdoor risks to indoor, leading to unintended exposure to sick building syndrome (SBS) symptoms. The aim of this study was to establish a comparative framework for the application of multiple logistic regression (classic regression) and credit scorecard models based on Logistic Regression (logit) and Decision Tree (DT) models in predicting the risk of developing SBS among citizens of Alborz province in Iran. The results of the classic regression indicated that the number of residents in the home (OR = 1.127, 95% CI: 1.017–1.760), were shown to have a significant association with total SBS symptoms. In the credit scorecard model utilizing the logit model, the variable with the greatest predictive significance for total SBS as well as general, and mucosal symptoms was work status. Furthermore, in the credit scorecard model based on DT model, the levels of PM2.5 and PM10, were the most significant predictors for the total SBS and dermal symptoms, respectively. As such, ozone (O3) emerged as the most influential factor in predicting both general and mucosal symptoms. As credit scoring, the greatest base score (109.86) was associated with mucosal symptoms, while the lowest base score (82.10) was associated with general symptoms. In conclusion, the DT model exhibited superior performance as a scoring model compared to the logit model and classic regression, as evidenced by its higher performance indexes. As such, only dermal symptoms exhibited mean scores higher than the base scores, suggesting that the overall quality of the indoor environments was generally poor.