Accurately predicting the compressive strength of concrete is crucial for quality control and structural safety assessment in civil engineering. However, traditional machine learning methods often struggle to capture the complex nonlinear interactions and strong feature correlations inherent in concrete mix design, thereby limiting their predictive accuracy and generalization ability. To address these challenges, this study proposes a deep learning model [AM-ResNet] that combines attention mechanisms with residual networks. The residual architecture effectively mitigates the vanishing gradient problem in deep networks, while the attention mechanism enhances the ability to identify key influencing factors by dynamically assigning feature weights. The experiments utilized a standard concrete compressive strength dataset from the UCI Machine Learning Repository, which contains 1,030 samples and 8 input variables. The study conducted a comprehensive, multidimensional comparative analysis of Support Vector Regression [SVR], Random Forest, and Backpropagation Neural Network models. The proposed AM-ResNet model achieved a coefficient of determination [R²] of 0.790, a root mean square error [RMSE] of 3.21 MPa, and a root mean square percentage error [RMSPE] of 8.9 %. Over 92 % of the prediction errors fell within the ±10 MPa range, meeting typical engineering quality control standards. Under 20 % input noise, AM-ResNet still maintains an R² value of 0.674, with a performance decline of only 14.7 %, significantly outperforming baseline models. Using only 10 % of the training samples, its R² value reaches 0.621, demonstrating exceptional data efficiency. Five-fold cross-validation confirmed its stability and demonstrated the lowest variance [σ = 0.021]. The proposed AM-ResNet framework provides a robust, accurate, and interpretable solution for concrete strength prediction, holding great potential for applications in intelligent quality control and material design within the civil engineering field.