The study developed a model to predict the risk of incipient pneumoconiosis using machine learning based on age and physical examination blood test results. Predictors were analyzed using LASSO and multiple logistic regression, while 9 models were tested: Logistic Regression, XGBoost, LightGBM, Random Forest, AdBoost, Gaussian Naïve Bayes, Multilayer Perceptron and SVM. Performance was evaluated by ROC, accuracy, sensitivity, specificity, PPV, NPV, F1 score, DCA, calibration curves and PR curves. 6 risk variables were identified: white blood cells (WBC), platelet distribution width (PDW), total bilirubin (TB), absolute neutrophil count (ANC), alanine aminotransferase (ALT) and aspartate aminotransferase (AST). The optimal model was SVM with good clinical applicability ratings. SHAP analysis explained the contribution of these variables to the progression of pneumoconiosis. The model has the potential to improve early detection in clinical practice.