Machine learning assessment of white blood cell counts in workers exposed to benzene: a historical cohort study

Abstract

To explore the fitting effect of the ARIMA, GM(1,1), and RANSAC model in the changes of white blood cells (WBC) in benzene-exposed workers, and select the optimal model to predict the WBC count of workers. Among 350 employees in an aerospace process manufacturing enterprise in Nanjing, workers with 10 years of benzene exposure were selected, and used Excel software to organize the WBC data, and the ARIMA model and RANSAC model were established by R software, and the GM(1, 1) model was established by DPS software, and the magnitude of the mean absolute percentage error (MAPE) of fitting three models to WBC counts was compared. The MAPE based on the ARIMA(2,1,2) model is 6.78%, the MAPE based on the GM(1,1) model is 5.19%, and the MAPE based on the RANSAC model is 6.37%, so the GM( 1,1) model was more suitable for fitting the trend of WBC counts in benzene exposed workers in this study. The GM(1,1) model is suitable for fitting WBC counts in a small sample size and can provide a short-term prediction of WBC counts in benzene-exposed workers and provide basic information for occupational health risk assessment of workers.

Keywords: ARIMA model; GM(1,1) model; Machine learning; RANSAC model·benzene exposed; WBC counts.