Understanding　and　predicting　the　physicochemical　properties　of　mixed　surfactant　systems　is　a　core　challenge　in　colloid　chemistry　and　industrial　applications,　due　to　their　diverse　types,　variable　ratios,　and　the　multifaceted　mixed　interaction　mechanisms.　This　work　innovatively　constructed　a　mixed　surfactant　dataset,　breaking　through　the　dual　limitations　of　data　scarcity　and　theoretical　simplifications　in　traditional　methods,　and　established　a　machine　learning　framework　guided　by　quantitative　structure-property　relationship　(QSPR).　By　theoretically　correcting　surface　tension　data,　the　XGBoost　model　achieved　prediction　accuracy　with　R2　of　0.9994　and　MSE　of　0.0676.　The　transfer　learning　strategy　was　combined　to　solve　the　generalization　challenge　across　different　concentrations　and　ratios.　Multi-scale　feature　importance　analysis　revealed　that　concentration,　ratio,　and　interaction　parameters,　as　introduced　descriptors,　significantly　affected　the　prediction　of　surface　tension　in　the　mixed　system.　This　research　provided　a　quantitative　basis　for　low　concentration,　high　efficiency　formulation　designed　and　offered　an　effective　and　cost-efficient　tool　for　mixed　surfactants　in　both　research　and　industry.