Abstract: Tunnel boring machine (TBM) operation monitoring is becoming increasingly sophisticated. However, the high computational cost restricts the deployment and application of data-driven technologies in tunnel projects. To solve this issue, an in-situ data screening method is proposed for the tunnel boring machine based on data clustering and global sensitivity analysis, aiming to reduce the computational cost of data mining and analysis tasks by sample size reduction and parameter selection, respectively. This method designs a new sample difference metric based on Euclidean distance and output variable variance. The Lagrange multiplier method is used to determine the respective weights of Euclidean distance and output variable variance. A solution strategy for the clustering objective function is derived, enabling the in-situ data screening in terms of sample size. Global sensitivity analysis is used to locate and select key parameters, enabling the in-situ data screening in terms of parameter number, ultimately forming a core dataset of TBM in-situ data. The proposed method is validated through the advance rate and geology prediction tasks of a tunnel boring machine. The experimental results indicate that the proposed sample size reduction method can greatly reduce the computational cost and maintain the prediction accuracy at the same time for both the advance rate and geology prediction tasks. The computational cost of the advance rate prediction model and the geological type prediction model is reduced by over 50.35%, and the prediction accuracy approaches that of the total samples (with a reduction of less than 5%).