<p>Real-time and accurate assessment of the Uniaxial Compressive Strength (UCS) of surrounding rock during Tunnel Boring Machine (TBM) tunneling is crucial for adjusting control parameters. However, existing methods face challenges in achieving the timely and accurate acquisition of UCS. Furthermore, there has yet to be research utilizing interpretable artificial intelligence to explore the relationship between tunneling parameters and UCS, which could provide new insights. To address these issues, this study proposes an interpretable artificial intelligence model to predict UCS based on TBM tunneling parameters. The proposed model is called the Joint Denoising and Weighted Interpretable Ensemble Model (JD-WIEM), which comprises data Joint Denoising (JD), a performance-driven weighted ensemble framework, and model interpretation. The JD approach utilizes the multi-level decomposition and collaborative processing strategies of multiple denoising methods to separate noise from different frequency components of complex signals. The performance-driven weighted ensemble framework leverages adaptive weighting to enhance the complementary performance strengths of heterogeneous base models, capturing data characteristics from multiple dimensions. Accumulated Local Effects and SHapley Additive exPlanations are used to determine the key features that predict UCS and to reveal the sensitivity of features to changes in UCS. Moreover, a model transfer strategy is proposed to enhance JD-WIEM’s applicability under diverse geological conditions. The model was validated using data from six water conveyance tunnels. The results indicate that JD-WIEM achieved a coefficient of determination of 0.9834, outperforming state-of-the-art methods. Using JD to process the tunneling data significantly improved the model’s prediction accuracy. From a global perspective, thrust was the most critical feature for predicting UCS. In addition, the thrust and Field Penetration Index (FPI) exhibited greater sensitivity to rapid increases in UCS within high-strength surrounding rocks; revolutions per minute was more sensitive to UCS rapid increases in low-strength surrounding rocks, and penetration showed higher sensitivity to rapid decreases in UCS. This study is vital for ensuring efficient TBM tunneling.</p><p><b>Highlights</b><UnorderedList Mark="Bullet"> <ItemContent> <p>A performance-driven Weighted Interpretable Ensemble Model framework is proposed to predict the Uniaxial Compressive Strength (UCS) of surrounding rock in TBM tunnels</p> </ItemContent> <ItemContent> <p>A Joint Denoising method based on multi-level decomposition and collaborative processing is proposed, and tunneling data processed with this method significantly enhances model accuracy</p> </ItemContent> <ItemContent> <p>A model transfer strategy is proposed, enhancing the applicability of the performance-driven weighted interpretable model under different geological conditions</p> </ItemContent> <ItemContent> <p>This study reveals the differential sensitivity of UCS variations to diverse tunneling parameters</p> </ItemContent> <ItemContent> <p>The proposed model achieves an R² of 0.9834, demonstrating superior performance compared to existing methods</p> </ItemContent> </UnorderedList></p>