The bulking factor of blasted muck piles in open-pit mines serves as a crucial metric for assessing blast performance. Precise estimation of muck pile volume is vital for optimizing blast parameters, controlling fragmentation effects, and planning subsequent excavation and haulage operations once the target excavation volume is established. To overcome limitations such as inefficiency, reliance on traditional volume-measurement approaches, and inadequate precision in small-sample predictions for blasted muck piles, this research develops an intelligent prediction model that leverages airborne LiDAR and machine learning algorithms. A lightweight UAV-LiDAR system for efficient, ground-control-independent acquisition of muck pile point cloud data was developed. Feature selection used the DeepSeek API to develop a hybrid model that integrates Pearson correlation analysis, ANOVA F-test, random forest feature importance assessment, and recursive feature elimination, ultimately extracting six critical factors from eight initial parameters. The GS-KCV-optimized Bayesian Ridge Regression model demonstrates predictive capability, achieving a test set R² of 0.76, a marked improvement over traditional approaches. The model reduces MAE to 19949.67 m³(26% decrease) and RMSE to 23020.80 m³(28% decrease), while effectively addressing the common limitation of local-optima convergence in small-sample scenarios. Field tests conducted with DJI M350 drones integrated with Zenmuse L₂ LiDAR systems verify the method′s ability to holistically optimize the operational process, spanning blast parameter collection, volume prediction, and performance assessment, providing a practical technological solution for intelligent mining transformation.