Study on a Bayes evaluation of the working ability of petroleum workers in the Karamay region, Xinjiang, China

OBJECTIVES: Use Bayes statistical methods to analyze the factors related to the working ability of petroleum workers in China and establish a predictive model for prediction so as to provide a reference for improving the working ability of petroleum workers. MATERIALS AND METHODS: The data come from the health questionnaire database of petroleum workers in the Karamay region, Xinjiang, China. The database contains the results of a health questionnaire survey conducted with 4,259 petroleum workers. We established an unsupervised Bayesian network, using Node-Force to analyze the dependencies between influencing factors, and established a supervised Bayesian network, using mutual information analysis methods (MI) to influence factors of oil workers’ work ability. We used the Bayesian target interpretation tree model to observe changes in the probability distribution of work ability classification under different conditions of important influencing factors. In addition, we established the Tree Augmented Naïve Bayes (TAN) prediction model to improve work ability, make predictions, and conduct an evaluation. RESULTS: (1) The unsupervised Bayesian network shows that there is a direct relationship between shoulder and neck musculoskeletal diseases, anxiety, working age, and work ability, (2) The supervised Bayesian network shows that anxiety, depression, shoulder and neck musculoskeletal diseases (Musculoskeletal Disorders, MSDs), low back musculoskeletal disorders (Musculoskeletal Disorders, MSDs), working years, age, occupational stress, and hypertension are relatively important factors that affect work ability. Other factors have a relative impact on work ability but are less important. CONCLUSION: Anxiety, depression, shoulder and neck MSDs, waist and back MSDs, and length of service are important influencing factors of work ability. The Tree Augmented Naïve Bayes prediction model has general performance in predicting workers’ work ability, and the Bayesian model needs to be deepened in subsequent research and a more appropriate forecasting method should be chosen.