Abstract: The complex and variable formation conditions pose a great challenge to safe and efficient production in coal mines. By establishing a drillability prediction model based on drilling data, intelligent perception of formation types can be achieved, providing geological information support for the drilling process.Addressing the issues of outliers and unbalanced sample sizes in coal mine drilling data, we employ the Local Outlier Factor (LOF) anomaly detection algorithm to remove abnormal drilling data. Additionally, we leverage a Generative Adversarial Network (GAN) to train on the original drilling samples, thereby obtaining generated drilling samples to construct a balanced drilling dataset.A method for predicting the drillability of a formation using a whale optimization algorithm-based kernel extreme learning machine (WOA-KELM) is proposed, which combines the drillability of the formation as an evaluation index. A prediction model based on the optimized WOA-KELM is constructed to predict the level of drillability of the formation.Through the actual drilling data in Huainan mining area, we obtained drilling parameters such as weight on bit (WOB), rotation speed, and penetration rate. After eliminating outliers using LOF, GAN was used to generate samples with high similarity to the original ones. The reliability of these generated drilling samples was verified through box plots, recognition rates, and scatter plot distributions.Using the generated sample balanced drilling dataset, a formation drillability prediction model was established through WOA-KELM for drilling process. The accuracy of formation drillability prediction reached about 98%, which is superior to recognition models such as SVM and KNN.The research results provide a reference for the adaptive drilling and intelligent perception technology in coal mines.