Magnesium-silicon-based　cement　(MSC)　is　a　novel　construction　material　composed　of　light-burned　magnesium　oxide　powder　and　various　siliceous　raw　materials,　offering　significant　low-carbon　advantages.　However,　the　types　and　elemental　compositions　of　raw　materials　introduce　substantial　uncertainty　into　the　mechanical　properties　of　MSC,　making　it　challenging　for　traditional　experimental　methods　to　predict　its　performance　accurately.　Machine　learning　(ML)　techniques　hold　considerable　benefits　in　dealing　with　complex　relationships.　This　study　aimed　to　use　ML　techniques　to　establish　a　predictive　model　for　the　compressive　strength　of　MSC　and　explore　the　impact　of　different　feature　parameters　on　the　compressive　strength　of　MSC　through　an　interpretable　model.　The　results　demonstrated　that　the　Extreme　Gradient　Boosting　(XGB)　model　excelled　in　prediction　accuracy.　Among　all　feature　parameters,　curing　age　was　the　most　crucial　factor　influencing　the　compressive　strength　of　MSC.　Element　content　analysis　indicated　that　there　was　a　nonlinear　relationship　between　the　compressive　strength　of　MSC　and　the　proportions　of　silicon　(P_Si)　and　magnesium　(P_Mg).　The　ideal　optimal　Mg/Si　ratio　was　approximately　0.76.　Furthermore,　the　analysis　of　interactions　between　elements　showed　that　nearly　balanced　contents　of　Mg　and　Si　produced　a　positive　interaction　effect　which　contributed　to　improving　the　mechanical　properties　of　MSC.　Al,　Ca,　and　Fe　exhibited　significant　variations　in　their　influence　on　the　Mg-Si　interaction　across　different　content　ranges,　presenting　clear　regional　dependencies　and　element-specific　effects.