ObjectiveWorkers exposed to dust for extended periods may experience varying degrees of cognitive impairment. However, limited research exists on the associated risk factors. This study aims to identify key variables using machine learning algorithms (ML) and develop a model to predict the occurrence of mild cognitive impairment in miners. Methods: A total of 1938 miners were included in the study. Univariate analysis and multivariable logistic regression were employed to identify independent risk factors for cognitive impairment among miners. The dataset was randomly divided into a training set and a validation set in an 8:2 ratio of 1550 and 388 individuals, respectively. An additional group of 351 miners was collected as a test set for cognitive impairment assessment. Seven machine learning algorithms, including XGBoost, Logistic Regression, Random Forest, Complement Naive Bayes, Multi-layer Perceptron, Support Vector Machine, and K-Nearest Neighbors, were used to establish a predictive model for mild cognitive impairment in the dust-exposed population, based on baseline characteristics of the workers. The predictive performance of the models was evaluated using the Area Under the Receiver Operating Characteristic Curve (AUC), and the XGBoost model was further explained using the Shapley Additive exPlanations (SHAP) package. Cognitive function assessments using rank sum tests were conducted to compare differences in cognitive domains between the mild cognitive impairment group and the normal group. Results: Univariate and multivariable logistic regression analyses revealed that education level, Age, Work years, SSRS (Self-Rating Scale for Stress), and HAMA (Hamilton Anxiety Rating Scale) were independent risk factors for cognitive impairment among dust-exposed workers. Comparative analysis of the performance of the seven machine learning algorithms demonstrated that XGBoost (training set: AUC=0.959, validation set: AUC=0.795) and Logistic Regression (training set: AUC=0.818, validation set: AUC=0.816) models exhibited superior predictive performance. Results from the test set showed that the AUC of the XGBoost model (AUC=0.913) outperformed the Logistic Regression model (AUC=0.778). Miners with mild cognitive impairment exhibited significant impairments (p<0.05) in visual-spatial abilities, attention, abstract thinking, and memory areas. Conclusion: Machine learning algorithms can predict the risk of cognitive impairment in this population, with the XGBoost algorithm showing the best performance. The developed model can guide the implementation of appropriate preventive measures for dust-exposed workers.