Prediction of microbial concentration in hospital indoor air based on gra-dient boosting decision tree model

Abstract:

Objective To explore the prediction of hospital indoor microbial concentration in air based on real-time indoor air environment monitoring data and machine learning algorithms. Methods Four locations in a hospital were selected as monitoring sampling points from May 23 to June 5, 2022. The "internet of things" sensor was used to monitor a variety of real-time air environment data. Air microbial concentration data collected at each point were matched, and the gradient boosting decision tree (GBDT) was used to predict real-time indoor microbial concentration in air. Five other common machine learning models were selected for comparison, including random forest (RF), decision tree (DT), k-nearest neighbor (KNN), linear regression (LR) and artificial neural network (ANN). The validity of the model was verified by the mean absolute error (MAE), root mean square error (RMSE) and mean absolute percentage error (MAPE). Results The MAPE value of GBDT model in the outpatient elevator room (point A), bronchoscopy room (point B), CT waiting area (point C), and nurses' station in the supply room (point D) were 22.49%, 36.28%, 29.34%, and 26.43%, respectively. The mean performance of the GBDT model was higher than that of other machine learning models at three sampling points and slightly lower than that of the ANN model at only one sampling point. The mean MAPE value of GBDT model at four sampling points was 28.64%, that is, the predicted value deviated from the actual value by 28.64%, indicating that GBDT model has good prediction results and the predicted value was within the available range. Conclusion The GBDT machine learning model based on real-time indoor air environment monitoring data can improve the prediction accuracy of indoor air microbial concentration in hospitals.