<p>Occupational noise hazards is a major health and safety issue in surface iron ore mining where heavy machinery continuously operates throughout the day. Most existing studies only describe noise conditions but do not provide reliable tools to predict when and where dangerous noise levels may occur. To address this gap, present study proposes a machine learning framework that predicts and classifies noise risk levels using field data from two operational surface iron ore mines. To build and validate this framework, noise levels across major operational zones such as crushers, shovels, dumpers, wheel loaders, and spotter locations were collected for three months. A total of 2,700 readings were recorded, with 450 kept aside for model testing. Each reading was labelled as “Little Risk” (250 samples), “Danger Limit” (173 samples), or “Warning Limit” (27 samples). Four machine learning models namely k-Nearest Neighbors (KNN), Gradient Boosting Regression (GBR), Random Forest (RF), and Support Vector Machine (SVM) were trained to predict these risk categories. During the field measurements peak noise levels recorded as 118.5 dBA near wheel loaders, and danger-limit conditions were frequently obtained near crushers in both selected iron ore mines. Among all models, SVM perform superior, with an F1 score of 0.96 and a Jaccard index of 0.91. The model performs well with high precision of 0.97 and recall value of 1.0 for recognising danger limit. The results show that machine learning can accurately identify high-risk noise conditions in surface mines, enabling early and targeted safety actions. The proposed system offers a practical tool for improving occupational noise management and supporting compliance with international safety standards.</p>