Machine Learning for Prediction the Severity of Restrictive Defect of Lung among Factory Workers

Abstract

Restrictive lung disease such as pneumoconiosis is the most common disease among people working in dusty environment such as mine and industry. The gold standard diagnosis for this disease is spirometry, which is used to evaluate both obstructive defect and restrictive defect. However, this tool has certain limitations such as high service costs, limited access to the device, and availability of specialists. These limitations impede early detection of this disease. The objective of this study is to use machine learning algorithms to help predicting the severity of restrictive defect among factory workers. Three severity classes considered are Normal, Mild, and Moderate or Severe. By using spirometry’s results and behavioral data among workers in a furniture factory in Thailand, six machine learning algorithms were developed, including Logistic Regression, Decision Tree, Random Forest, Gradient Boosting, XGBoost and Support Vector Machine (SVM). The best model is Random Forest with Synthetic Minority Oversampling (SMOTE) to deal with imbalance class and Recursive Feature Elimination (RFE) to select most important features. The important features are weight, height, age, education, hours of work, smoking, mask wearing at the f1-score = 0.746, precision = 0.743, recall = 0.756, and accuracy = 0.75. The model was deployed through a web application for ease of use.  The application was used among the factory workers for early screen of the disease. The users were satisfied with the application for its effectiveness, ease of use, time, and cost savings. The tool may increase the chances for early detection and assessment of restrictive lung disease and reduce the burden of future diseases.