Abstract: In the field of construction and municipal engineering, the performance evaluation of existing concrete structures, especially the prediction of compressive strength, has become a hot topic. Non destructive testing techniques (NDT) represented by ultrasonic pulse velocity (UPV) and rebound number (RN) testing are widely used for predicting the compressive strength of concrete structures. However, most existing methods ignore the factors of concrete mix design, resulting in insufficient model generalization ability and difficulty in adapting to the diverse concrete composition in practical engineering. This study established two machine learning models, Support Vector Regression (SVR) and Backpropagation Neural Network (BPNN), based on non-destructive testing data of concrete with different mix proportions. The models were trained on a dataset of 180 different mix proportions, and the performance of the models was further evaluated using machine learning evaluation indicators. The prediction results show that, without considering the concrete mix design, the determination coefficients R² of the concrete compressive strength test set corresponding to the SVR model and BPNN prediction model are 0.860 and 0.867, respectively. When considering the design of concrete mix proportion, the determination coefficient R² of the SVR model prediction results is 95%, and the determination coefficient R² of the BPNN prediction model is 92%. Moreover, the SVR model outperforms the BPNN model in all evaluation indicators of concrete compressive strength prediction results. This study provides a high-precision and non-destructive technical means for evaluating the strength of existing concrete structures, which has important engineering application value.