Active learning for optimal feature combination selection in machine learning of TBM operational data

Although extensive operational data from hard rock tunnel boring machines (TBMs) enable an extraction of numerous features for different prediction tasks using machine learning (ML), incorporating all TBM data features for ML may lead to high computation costs and potentially adverse model performance. This renders a significant challenge in ML of TBM data, i.e., how to determine an optimal combination of TBM features for improving model fidelity and efficiency. To address this challenge, this paper proposes an active learning method that intelligently selects superior feature combinations by evaluating performance of models developed from carefully selected subsets of features and adaptively changing the feature combinations. The proposed method is illustrated through a water conveyance tunnel project in China. It offers a robust framework for streamlining data processing while ensuring reliable predictions. In future studies, generalization capability of the proposed method will be further validated across more tunnel projects. • An active learning method is proposed for optimal selection of TBM feature combination. • TBM feature combinations are effectively described using a vector representation. • Combining GPR with EI to guide the search for a potential superior TBM feature combination. • The optimal TBM feature combination improves fidelity and efficiency of the model.