<p>To improve the accuracy of ball mill load state identification, this study proposes a subdivision load identification method based on multi-domain feature extraction with UMAP-BOXGBoost. Firstly, it was verified that both ball-charge volume ratio and material-ball volume ratio were important indicators affecting the ball mill load through the analysis of the particle size after grinding, and the ball mill load was subdivided according to the low, medium, and high combinations of the two indicators in a total of nine states. Secondly, on the basis of extracting features based on time domain, frequency domain, and entropy domain, the signal is converted into viewable extracted signal graph domain features, and the four domain features are combined to establish a richer set of ball mill subdivision load features. Finally, the XGBoost algorithm is used to achieve efficient training of the feature set, and Bayesian optimization is used to find the optimal hyper-parameters of XGBoost in order to solve the problem of the XGBoost parameters being difficult to set correctly, and the data are mapped to low dimensions by the UMAP algorithm and then input into the recognition model in order to improve the recognition accuracy. Experiments show that under multi-domain feature extraction, UMAP-BOXGBoost has an accuracy of 98.85%, with higher robustness and classification accuracy, which is more effective for ball mill segmentation load.</p>