Sulfate attack is a critical factor affecting concrete durability, and accurate prediction of its detrimental effect on concrete compressive strength is essential for structural safety assessment. A machine learning-based approach was proposed to predict the compressive strength of concrete subjected to sulfate corrosion. A total of 157 experimental datasets on the compressive strength of concrete under sulfate attack were collected from existing domestic and international literature. The data were divided into training and test sets with a ratio of 80% and 20%. RF (random forest) and BPNN (BP neural network) models were developed to predict the compressive strength of concrete after sulfate corrosion. The performance of the models was evaluated using three indexes: MAE (mean absolute error), RMSE (root mean squared error), and the coefficient of determination (R²). To further improve prediction accuracy, feature importance, Pearson correlation coefficient, and a combination of both methods were applied for optimization. Based on the optimal optimization method, the model was further interpreted. The results show that both models provide accurate predictions of compressive strength after sulfate corrosion with the RF model demonstrating superior performance compared to the BPNN model. Optimization analysis reveales that the combination of feature importance and Pearson correlation coefficient yield the best predictive performance. Based on the best optimization method, the influence of different factors on 28 d compressive strength of eroded concrete was analyzed. The prediction models with variation of 28-day compressive strength, erosion time and sodium sulfate concentration were established. This study provides a reference for the application of big data techniques in predicting the compressive strength of concrete in sulfate corrosion environments.