Predicting pneumoconiosis risk in coal workers using artificial neural networks

Abstract

Objectives: This study aimed to create a model to predict pneumoconiosis risk in coal workers using Methods: An ANN-based model was developed using the health records of a population of coal workers (all men). Input neurons comprised current age, year the worker began his employment, occupational category, the number of days spent working underground, the total days spent working, the duration of employment in working underground (i.e., in a so-called group 1 job), and smoking status. Output neurons comprised the states of having pneumoconiosis and being free of pneumoconiosis. Results: The study found that an ANN model incorporating the variables age, the duration of employment in a group 1 job, the number of days spent working underground, year the worker began his employment, the total days spent working, smoking status, and occupational category can be used to estimate pneumoconiosis risk. The model's success rate was 95.3%; sensitivity was 90.3%, and specificity was 96.5%. The most influential input variable for pneumoconiosis was age, followed by the duration of employment in a group 1 job. Conclusion: Predicting pneumoconiosis risk in coal workers provides great advantages for strategically monitoring miners and developing preventive health programs. Artificial neural network models should be developed, integrated into occupational medicine practice, and used to evaluate workers' health status.