Abstract: Based on the damage depth of coal seam floor prediction method and theory, according to the cases of the coal seam floor damage depth collected on site in mining fields, it is analyzed that the mining depth, coal seam pitch, mining thickness, working face length, the floor damage resistant ability and the existence of fault or fracture are main factors influencing the damage depth of the coal seam floor. To overcome the overfitting problem for the artificial neural network method, a novel method for predicting the damage depth of coal seam floor by Least-Squares Support Vector Machines (LS-SVM) is proposed in this paper whose hyper-parameter selection is presented based on the Particle Swarm Optimization (PSO). As it is difficult to determine the machine parameters of LS-SVM and the prediction accuracy is not high, the PSO is applied for its high convergence speed and global optimization ability, this paper optimizes the penalty factor and kernel function parameters of LS-SVM model to avoid the blindness of the manual parameter choice and to improve the training speed and the generalization ability of the prediction model. Statistical data of the mining-induced coal seam floor damage depth were calculated for the main mining areas in China by applying the LS-SVM prediction model and the results indicate that the value predicted by the model coincides with the actual measured data and is more reliable than the value calculated by the empirical formula. The model has not only a reliable theoretical basis but also good application value.