In order to clarify the deterioration law of high-temperature compressive strength of self-compacting concrete, an integrated machine learning model was employed in this study to construct an interpretable prediction framework for forecasting the residual compressive strength of self-compacting concrete after high-temperature exposure, whereby the intrinsic mechanism of strength loss was revealed. The results show that the XGBoost model optimized by the sparrow search algorithm exhibits outstanding predictive performance for the residual compressive strength of self-compacting concrete following high-temperature treatment, with a prediction accuracy of 98.0%. The overall feature importance analysis demonstrates that fire exposure temperature serves as the dominant factor affecting the high-temperature residual compressive strength, followed by fine aggregate dosage and water-binder ratio. The feature interaction network analysis indicates that fire exposure temperature presents negative interactive effects with both water-binder ratio and fine aggregate dosage, while a positive interactive correlation is found between fire exposure temperature and supplementary cementitious material dosage. The addition of supplementary cementitious materials can effectively improve the high-temperature compressive performance of self-compacting concrete. Furthermore, an online intelligent prediction tool was developed in this study. This tool enables the rapid and accurate prediction of high-temperature compressive strength and the visual characterization of core driving factors for high-temperature strength loss as well as their corresponding contribution degrees at both global and individual scales. The research outcomes can provide solid theoretical support and intelligent decision-making guidance for the post-fire performance evaluation and precise remediation of self-compacting concrete components.